[GitHub] [incubator-mxnet] haojin2 commented on a change in pull request #15493: [numpy] numpy einsum
haojin2 commented on a change in pull request #15493: [numpy] numpy einsum
URL: https://github.com/apache/incubator-mxnet/pull/15493#discussion_r302276084
##
File path: src/operator/numpy/np_einsum_op-inl.h
##
@@ -0,0 +1,772 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_einsum_op-inl.h
+ * \brief Function definition of numpy-compatible einsum operator
+ */
+
+#ifndef MXNET_OPERATOR_NUMPY_NP_EINSUM_OP_INL_H_
+#define MXNET_OPERATOR_NUMPY_NP_EINSUM_OP_INL_H_
+
+#include
+#include
+#include
+#include "../../common/static_array.h"
+#include "../mxnet_op.h"
+#include "../operator_common.h"
+#include "../mshadow_op.h"
+#include "../elemwise_op_common.h"
+
+
+namespace mxnet {
+namespace op {
+
+#define NPY_MAXDIMS 32
+#define NPY_MAXARGS 32
+
+
+inline TShape get_stride(const TShape& shape) {
+ int ndim = shape.ndim(), prod = 1;
+ TShape stride = TShape(ndim, -1);
+ for (int i = ndim - 1; i >= 0; i--) {
+stride[i] = shape[i] > 1 ? prod : 0;
+prod = prod * shape[i];
+ }
+ return stride;
+}
+
+
+inline TShape pad(const TShape& shape, int odim) {
+ int ndim = shape.ndim();
+ CHECK_GE(odim, ndim);
+ TShape ret(odim, 1);
+ for (int idim = 0; idim < ndim; ++idim) {
+ret[idim] = shape[idim];
+ }
+ return ret;
+}
+
+
Review comment:
one less blank line here
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] roywei commented on issue #15431: [MXNet 1.5.0.rc2] Issues with asnumpy() method
roywei commented on issue #15431: [MXNet 1.5.0.rc2] Issues with asnumpy() method
URL:
https://github.com/apache/incubator-mxnet/issues/15431#issuecomment-510236197
@Wallart your code is running fine on my machine. I'm testing on[ LJ-Speech
Dataset](https://keithito.com/LJ-Speech-Dataset/)
added a few lines in `main` to format the text files, and everything is
plotted, and `asnumpy()` works fine. Could you try without Docker?
```
if __name__ == '__main__':
logging.basicConfig()
logging.getLogger().setLevel(logging.INFO)
params = {
'max_wav_value': 32768.0, # for 16 bits files
'sampling_rate': 22050,
'filter_length': 1024,
'hop_length': 256,
'win_length': 1024,
'n_mel_channels': 80,
'mel_fmin': 0.0,
'mel_fmax': 8000.0
}
with open('~/Downloads/LJSpeech-1.1/metadata.csv', encoding='utf-8') as
f:
for line in f:
record = line.split('|')
file_name = record[0] + '.txt'
content = record[2]
with open('~/Downloads/LJSpeech-1.1/wavs/' + file_name, 'w',
encoding='utf-8') as text_output:
text_output.write(content)
french = WavDataset('~/Downloads/LJSpeech-1.1/wavs', text_to_sequence,
**params)
assert type(french[0]) == tuple
```
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] roywei edited a comment on issue #15431: [MXNet 1.5.0.rc2] Issues with asnumpy() method
roywei edited a comment on issue #15431: [MXNet 1.5.0.rc2] Issues with
asnumpy() method
URL:
https://github.com/apache/incubator-mxnet/issues/15431#issuecomment-510236197
@Wallart your code is running fine on my machine. I'm testing on[ LJ-Speech
Dataset](https://keithito.com/LJ-Speech-Dataset/)
added a few lines in `main` to format the text files, and everything is
plotted, and `asnumpy()` works fine. Could you try without Docker?
```
if __name__ == '__main__':
logging.basicConfig()
logging.getLogger().setLevel(logging.INFO)
params = {
'max_wav_value': 32768.0, # for 16 bits files
'sampling_rate': 22050,
'filter_length': 1024,
'hop_length': 256,
'win_length': 1024,
'n_mel_channels': 80,
'mel_fmin': 0.0,
'mel_fmax': 8000.0
}
with open('~/Downloads/LJSpeech-1.1/metadata.csv', encoding='utf-8') as
f:
for line in f:
record = line.split('|')
file_name = record[0] + '.txt'
content = record[2]
with open('~/Downloads/LJSpeech-1.1/wavs/' + file_name, 'w',
encoding='utf-8') as text_output:
text_output.write(content)
french = WavDataset('~/Downloads/LJSpeech-1.1/wavs', text_to_sequence,
**params)
assert type(french[0]) == tuple
```
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] haojin2 commented on a change in pull request #15493: [numpy] numpy einsum
haojin2 commented on a change in pull request #15493: [numpy] numpy einsum
URL: https://github.com/apache/incubator-mxnet/pull/15493#discussion_r302284030
##
File path: src/operator/numpy/np_einsum_op.cc
##
@@ -0,0 +1,192 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_einsum_op.cc
+ * \brief CPU Implementation of numpy-compatible einsum
+ */
+
+#include "./np_einsum_op-inl.h"
+#include
+#include
+
+namespace mxnet {
+namespace op {
+
+inline bool NumpyEinsumShape(const nnvm::NodeAttrs& attrs,
+ mxnet::ShapeVector *in_attrs,
+ mxnet::ShapeVector *out_attrs) {
+ const NumpyEinsumParam ¶m = nnvm::get(attrs.parsed);
+ const char* subscripts = param.subscripts.c_str();
+ int num_args = param.num_args;
+ CHECK_EQ(in_attrs->size(), num_args);
+ CHECK_EQ(out_attrs->size(), 1U);
+ for (int i = 0; i < num_args; i++) {
+if (!shape_is_known(in_attrs->at(i))) {
+ return false;
+}
+ }
+
+ int iop, label, min_label = 127, max_label = 0;
+ int nop = num_args;
+ char label_counts[128];
+ int label_size[128], max_broadcast = -1;
+ char op_labels[NPY_MAXARGS][NPY_MAXDIMS];
+ char output_labels[NPY_MAXDIMS];
+ int idim, ndim_output, ndim_broadcast;
+
+ /* Parse the subscripts string into label_counts and op_labels */
+ memset(label_counts, 0, sizeof(label_counts));
+ for (iop = 0; iop < nop; ++iop) {
+int length = static_cast(strcspn(subscripts, ",-"));
+CHECK(!(iop == nop-1 && subscripts[length] == ','))
+ << "more operands provided to einstein sum function "
+ "than specified in the subscripts string";
+CHECK(!(iop < nop-1 && subscripts[length] != ','))
+ << "fewer operands provided to einstein sum function "
+ "than specified in the subscripts string";
+CHECK_GE(parse_operand_subscripts(subscripts, length,
+ in_attrs->at(iop).ndim(),
+ iop, op_labels[iop], label_counts,
+ &min_label, &max_label), 0);
+
+/* Move subscripts to the start of the labels for the next op */
+subscripts += length;
+if (iop < nop - 1) {
+ subscripts++;
+}
+ }
+
+ /*
+ * Find the number of broadcast dimensions, which is the maximum
+ * number of labels == 0 in an op_labels array.
+ */
+ ndim_broadcast = 0;
+ for (iop = 0; iop < nop; ++iop) {
+int count_zeros = 0;
+int ndim;
+char *labels = op_labels[iop];
+ndim = in_attrs->at(iop).ndim();
+for (idim = 0; idim < ndim; ++idim) {
+ if (labels[idim] == 0) {
+++count_zeros;
+ } else if (labels[idim] > 0) {
+label_size[static_cast(labels[idim])] = in_attrs->at(iop)[idim];
+ }
+}
+if (count_zeros > ndim_broadcast) {
+ ndim_broadcast = count_zeros;
+ max_broadcast = iop;
+}
+ }
+
+ /*
+ * If there is no output signature, fill output_labels and ndim_output
+ * using each label that appeared once, in alphabetical order.
+ */
+ if (subscripts[0] == '\0') {
+/* If no output was specified, always broadcast left, as usual. */
+for (ndim_output = 0; ndim_output < ndim_broadcast; ++ndim_output) {
+ output_labels[ndim_output] = 0;
+}
+for (label = min_label; label <= max_label; ++label) {
+ if (label_counts[label] == 1) {
+CHECK(ndim_output < NPY_MAXDIMS)
+ << "einstein sum subscript string has too many "
+ << "distinct labels";
+output_labels[ndim_output++] = label;
+ }
+}
+ } else {
+CHECK(subscripts[0] == '-' && subscripts[1] == '>')
+ << "einstein sum subscript string does not "
+ << "contain proper '->' output specified";
+subscripts += 2;
+
+/* Parse the output subscript string. */
+ndim_output = parse_output_subscripts(subscripts, strlen(subscripts),
+ ndim_broadcast, label_counts,
+ output_labels);
+CHECK_GE(ndim_output, 0);
+ }
+
+ // std::cout << "output_labels" << std::endl;
+ // for (int i = 0; i < ndim_output; ++i) {
+ // std::cout << output_labels[i] << " ";
+
[GitHub] [incubator-mxnet] lanking520 merged pull request #15500: fix the bug on Scala Sparse
lanking520 merged pull request #15500: fix the bug on Scala Sparse URL: https://github.com/apache/incubator-mxnet/pull/15500 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] IvyBazan commented on issue #15454: Julia docs
IvyBazan commented on issue #15454: Julia docs URL: https://github.com/apache/incubator-mxnet/pull/15454#issuecomment-510260831 LGTM! Verified Julia installation and doc building process on an Ubuntu EC2 instance. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] Zha0q1 commented on issue #15490: [WIP] Utility to help developers debug operators: Tensor Inspector
Zha0q1 commented on issue #15490: [WIP] Utility to help developers debug operators: Tensor Inspector URL: https://github.com/apache/incubator-mxnet/pull/15490#issuecomment-510275321 > I could probably rewrite build_checker() (https://github.com/apache/incubator-mxnet/pull/15490/files#diff-9e7d5c2420ecc900c4a85a3c35d91bffR446) with macro to suppress warnings when DType is not floating point. But it's not causing any issues because in such cases the control flow wouldn't go by the undefined operation. Will change if have time. After experimenting for several hours today, I think there is not a good way to breach at compile time to suppress the warnings. Macro apparently did not work. I also tried simulating C++17 style `constexpr if` with templated structs, which is an idea proposed by https://baptiste-wicht.com/posts/2015/07/simulate-static_if-with-c11c14.html, but that method was compiler specific and did not work either. Having no warnings when compiling is nice, but given that this is a developer's tool and that behavior-wise the code has no issue, I think I will leave the warnings. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya opened a new issue #15506: Improving error message
ChaiBapchya opened a new issue #15506: Improving error message URL: https://github.com/apache/incubator-mxnet/issues/15506 Source build on Ubuntu (EC2) I ran into this error: Thought this error message could be improved - ``` src/operator/quantization/./.././../common/../operator/mxnet_op.h(772): catastrophic error: error while writing generated C file: No space left on device 1 catastrophic error detected in the compilation of "/tmp/tmpxft_32f8_-16_quantize.compute_35.cpp1.ii". Compilation terminated. : fatal error: when writing output to : No space left on device compilation terminated. Makefile:535: recipe for target 'build/src/operator/nn/layer_norm_gpu.o' failed make: *** [build/src/operator/nn/layer_norm_gpu.o] Error 1 Makefile:535: recipe for target 'build/src/operator/quantization/quantize_gpu.o' failed make: *** [build/src/operator/quantization/quantize_gpu.o] Error 1 : fatal error: when writing output to : No such file or directory compilation terminated. ``` This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15506: Improving error message
mxnet-label-bot commented on issue #15506: Improving error message URL: https://github.com/apache/incubator-mxnet/issues/15506#issuecomment-510279254 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #15506: Improving error message
ChaiBapchya commented on issue #15506: Improving error message URL: https://github.com/apache/incubator-mxnet/issues/15506#issuecomment-510279558 @mxnet-label-bot add [Installation, Unclear Error/Doc, Build] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zhly0 commented on issue #15131: doc consistence
zhly0 commented on issue #15131: doc consistence URL: https://github.com/apache/incubator-mxnet/issues/15131#issuecomment-510288738 Thanks！ This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zhly0 closed issue #15131: doc consistence
zhly0 closed issue #15131: doc consistence URL: https://github.com/apache/incubator-mxnet/issues/15131 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zixuanweeei commented on a change in pull request #15497: Independent gradients requests check with respect to weights and bias of convolution
zixuanweeei commented on a change in pull request #15497: Independent gradients
requests check with respect to weights and bias of convolution
URL: https://github.com/apache/incubator-mxnet/pull/15497#discussion_r302333454
##
File path: src/operator/nn/mkldnn/mkldnn_convolution.cc
##
@@ -662,21 +662,21 @@ void MKLDNNConvolutionBackward(const nnvm::NodeAttrs&
attrs, const OpContext &ct
in_grad[conv::kWeight],
convBwdWeight.bwdWeights_pd.diff_weights_primitive_desc(),
req[conv::kWeight]);
-mkldnn_output_t in_grad_bias;
-if (param.no_bias) {
- convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second);
- MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
-} else {
- in_grad_bias = CreateMKLDNNMem(
+
+if (!param.no_bias && req[conv::kBias]) {
+ auto in_grad_bias = CreateMKLDNNMem(
in_grad[conv::kBias],
convBwdWeight.bwdWeights_pd.diff_bias_primitive_desc(),
req[conv::kBias]);
convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second, *in_grad_bias.second);
+ *in_grad_weight.second, *in_grad_bias.second);
MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
CommitOutput(in_grad[conv::kBias], in_grad_bias);
+} else {
Review comment:
Sure. I see. There is unnecessary primitive registration without the check
enabled. Thanks.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] Zha0q1 commented on issue #15490: Utility to help developers debug operators: Tensor Inspector
Zha0q1 commented on issue #15490: Utility to help developers debug operators: Tensor Inspector URL: https://github.com/apache/incubator-mxnet/pull/15490#issuecomment-510294519 @apeforest @anirudh2290 @access2rohit @larroy @sandeep-krishnamurthy , Hi guys, I completed this PR and it's ready for review. The only thing that I do not have yet is support for CSR NDArrays because I want to ask your opinion. I think I probably do not need to generate the whole matrix. Having something like this in my interactive print should be good enough? (this is the existing test::print)  This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] Zha0q1 edited a comment on issue #15490: Utility to help developers debug operators: Tensor Inspector
Zha0q1 edited a comment on issue #15490: Utility to help developers debug operators: Tensor Inspector URL: https://github.com/apache/incubator-mxnet/pull/15490#issuecomment-510294519 @apeforest @anirudh2290 @access2rohit @larroy @sandeep-krishnamurthy , Hi guys, I completed this PR and it's ready for review. The only thing that I do not have yet is support for CSR NDArrays because I want to ask your opinion. I think I probably do not need to generate the whole matrix. Having something like this in my interactive print should be good enough? (this is in the existing test::print)  This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] KellenSunderland commented on issue #15399: Add unit tests for TensorRT integration and fix some bugs
KellenSunderland commented on issue #15399: Add unit tests for TensorRT integration and fix some bugs URL: https://github.com/apache/incubator-mxnet/pull/15399#issuecomment-510294845 Looks like CI caught a few issues. For example http://jenkins.mxnet-ci.amazon-ml.com/blue/organizations/jenkins/mxnet-validation%2Funix-gpu/detail/PR-15399/1/pipeline seems like it should be relevant to this PR. I'll have a look to see if there's anything else that jumps out at me. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] KellenSunderland commented on a change in pull request #15399: Add unit tests for TensorRT integration and fix some bugs
KellenSunderland commented on a change in pull request #15399: Add unit tests for TensorRT integration and fix some bugs URL: https://github.com/apache/incubator-mxnet/pull/15399#discussion_r302336971 ## File path: src/operator/subgraph/tensorrt/nnvm_to_onnx.cc ## @@ -157,6 +157,12 @@ std::string ConvertNnvmGraphToOnnx( return serialized_onnx_graph; } +void ConvertIdentity(NodeProto* node_proto, const NodeAttrs& attrs, Review comment: Any idea if TRT actually optimizes this out? I've seen this in a few prod services :-/ This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] KellenSunderland commented on a change in pull request #15399: Add unit tests for TensorRT integration and fix some bugs
KellenSunderland commented on a change in pull request #15399: Add unit tests
for TensorRT integration and fix some bugs
URL: https://github.com/apache/incubator-mxnet/pull/15399#discussion_r302337630
##
File path: src/operator/subgraph/tensorrt/tensorrt-inl.h
##
@@ -109,13 +111,70 @@ class TensorrtSelector : public SubgraphSelector {
bool isTRTCompatible(const nnvm::Node &n) {
const std::string op_name = n.op()->name;
+if (op_name == "FullyConnected") {
+ const auto& param = nnvm::get(n.attrs.parsed);
+ return !param.no_bias;
+}
+
if (op_name == "Pooling") {
- return (n.attrs.dict.at("pool_type") == "avg" ||
- n.attrs.dict.at("pool_type") == "max");
+ const auto& param = nnvm::get(n.attrs.parsed);
+ if (param.layout.has_value()) {
+if (param.layout.value() == mshadow::kNHWC) {
+ LOG(INFO) << "Warning: NHWC layout (node: " << n.attrs.name
+<< ") is not supported by TensorRT";
+ return false;
+} else if (param.layout.value() == mshadow::kNDHWC) {
+ LOG(INFO) << "Warning: NDHWC layout (node: " << n.attrs.name
+<< ") is not supported by TensorRT";
+ return false;
+}
+ }
+ if (param.pooling_convention != pool_enum::kValid && !param.global_pool)
+return false;
+ if (param.pool_type == pool_enum::kAvgPooling) {
+if ((!param.global_pool) &&
+(!param.count_include_pad.has_value() ||
param.count_include_pad.value()))
+ return false;
+return true;
+ } else if (param.pool_type == pool_enum::kMaxPooling) {
+return true;
+ } else {
+return false;
+ }
}
-if (unconditionalTRTops.count(op_name)) {
- return true;
+if (op_name == "Convolution") {
+ const auto& param = nnvm::get(n.attrs.parsed);
+ if (!param.layout.has_value())
+return true;
+ switch (param.layout.value()) {
+case mshadow::kNCHW:
+case mshadow::kNCW:
+case mshadow::kNCDHW:
+ return true;
+case mshadow::kNHWC:
+ LOG(INFO) << "Warning: NHWC layout (node: " << n.attrs.name
+<< ") is not supported by TensorRT";
+ return false;
+case mshadow::kNDHWC:
+ LOG(INFO) << "Warning: NDHWC layout (node: " << n.attrs.name
+<< ") is not supported by TensorRT";
+ return false;
+default:
+ LOG(INFO) << "Warning: Layout (node: " << n.attrs.name
+<< ") is unknown (so unsupported by TensorRT)";
+ return false;
+ }
+}
+
+if (op_name == "Concat") {
+ const auto& param = nnvm::get(n.attrs.parsed);
+ return (param.dim != 0);
+}
+
+if (op_name == "Dropout") {
Review comment:
Again, will TensorRT optimize this out? We don't want it at inference time
right?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] KellenSunderland commented on a change in pull request #15399: Add unit tests for TensorRT integration and fix some bugs
KellenSunderland commented on a change in pull request #15399: Add unit tests
for TensorRT integration and fix some bugs
URL: https://github.com/apache/incubator-mxnet/pull/15399#discussion_r302338157
##
File path: src/operator/subgraph/tensorrt/tensorrt-inl.h
##
@@ -180,6 +253,17 @@ class TensorrtProperty : public SubgraphProperty {
n->attrs.name = "TensorRT" + std::to_string(subgraph_id);
n->attrs.op = Op::Get("_TensorRT");
CHECK(n->attrs.op);
+// prevent using Gamma value if using fix_gamma on BatchNorm
+DFSVisit(new_sym.outputs, [&n](const nnvm::NodePtr& node) {
+ if (node->op() == Op::Get("BatchNorm")) {
Review comment:
Why not just check for FixGamma = true during nnvm -> Onnx conversion and
set gamma to 0 if it's true?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] KellenSunderland commented on a change in pull request #15399: Add unit tests for TensorRT integration and fix some bugs
KellenSunderland commented on a change in pull request #15399: Add unit tests for TensorRT integration and fix some bugs URL: https://github.com/apache/incubator-mxnet/pull/15399#discussion_r302338438 ## File path: tests/python/gpu/test_tensorrt.py ## @@ -0,0 +1,437 @@ +# Licensed to the Apache Software Foundation (ASF) under one Review comment: Super helpful tests, thanks Clement. It's ok to merge in a subset of these at a time, but I'd try to avoid pushing commented out code / too many todos upstream. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] KellenSunderland commented on a change in pull request #15449: cuda/cuDNN lib version checking. Force cuDNN v7 usage.
KellenSunderland commented on a change in pull request #15449: cuda/cuDNN lib
version checking. Force cuDNN v7 usage.
URL: https://github.com/apache/incubator-mxnet/pull/15449#discussion_r302339983
##
File path: src/common/cuda_utils.cc
##
@@ -0,0 +1,108 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * Copyright (c) 2019 by Contributors
+ * \file cuda_utils.cc
+ * \brief CUDA debugging utilities.
+ */
+
+#include
+#include "cuda_utils.h"
+
+#if MXNET_USE_CUDA == 1
+
+namespace mxnet {
+namespace common {
+namespace cuda {
+
+// The oldest version of cuda used in upstream MXNet CI testing, both for unix
and windows.
+// Users that have rebuilt MXNet against older versions will we advised with a
warning to upgrade
+// their systems to match the CI level. Minimally, users should rerun the CI
locally.
+#if defined(_MSC_VER)
+#define MXNET_CI_OLDEST_CUDA_VERSION 9020
+#else
+#define MXNET_CI_OLDEST_CUDA_VERSION 1
+#endif
+
+// Dynamic init here will emit a warning if runtime and compile-time cuda lib
versions mismatch.
+// Also if the user has recompiled their source to a version no longer tested
by upstream CI.
+bool cuda_version_check_performed = []() {
+ // Don't bother with checks if there are no GPUs visible (e.g. with
CUDA_VISIBLE_DEVICES="")
+ if (dmlc::GetEnv("MXNET_CUDA_VERSION_CHECKING", true) &&
Context::GetGPUCount() > 0) {
+int linkedAgainstCudaVersion = 0;
+CUDA_CALL(cudaRuntimeGetVersion(&linkedAgainstCudaVersion));
+if (linkedAgainstCudaVersion != CUDA_VERSION)
+ LOG(WARNING) << "cuda library mismatch: linked-against version " <<
linkedAgainstCudaVersion
Review comment:
Just want to make sure I'm understanding this one. If a user runs with CUDA
10.2, but the library was linked against 10.1 would this issue a warning? I
tend to do that fairly often, is it against best practices?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] tingying2020 opened a new pull request #15507: np_around
tingying2020 opened a new pull request #15507: np_around URL: https://github.com/apache/incubator-mxnet/pull/15507 ## Description ## (Brief description on what this PR is about) ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [x] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ## Comments ## np_around do not support `float16` @haojin2 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] xidulu commented on a change in pull request #15495: [Numpy] Added operator logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
xidulu commented on a change in pull request #15495: [Numpy] Added operator
logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15495#discussion_r302343715
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cc
##
@@ -186,6 +212,21 @@
MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_power_scalar)
.set_attr("FCompute", BinaryScalarOp::Compute)
.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_power_scalar"});
+MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_logaddexp_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute)
+.set_attr("Fgradient",
ElemwiseGradUseIn{"_backward_logaddexp_scalar"});
+
+
+MXNET_OPERATOR_REGISTER_BINARY(_backward_logaddexp_scalar)
+.add_argument("scalar", "float", "scalar value")
+.set_attr_parser([](NodeAttrs *attrs) { attrs->parsed =
std::stod(attrs->dict["scalar"]); })
+.set_attr("FCompute", BinaryScalarOp::Backward<
+ cpu, mshadow_op::logadd_left>);
+
+
+
+
+
Review comment:
blank lines removed, thanks for reviewing
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] xidulu commented on a change in pull request #15495: [Numpy] Added operator logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
xidulu commented on a change in pull request #15495: [Numpy] Added operator
logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15495#discussion_r302343715
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cc
##
@@ -186,6 +212,21 @@
MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_power_scalar)
.set_attr("FCompute", BinaryScalarOp::Compute)
.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_power_scalar"});
+MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_logaddexp_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute)
+.set_attr("Fgradient",
ElemwiseGradUseIn{"_backward_logaddexp_scalar"});
+
+
+MXNET_OPERATOR_REGISTER_BINARY(_backward_logaddexp_scalar)
+.add_argument("scalar", "float", "scalar value")
+.set_attr_parser([](NodeAttrs *attrs) { attrs->parsed =
std::stod(attrs->dict["scalar"]); })
+.set_attr("FCompute", BinaryScalarOp::Backward<
+ cpu, mshadow_op::logadd_left>);
+
+
+
+
+
Review comment:
blank lines removed, thanks for reviewing
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] xidulu commented on a change in pull request #15495: [Numpy] Added operator logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
xidulu commented on a change in pull request #15495: [Numpy] Added operator
logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15495#discussion_r302343855
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cc
##
@@ -186,6 +212,21 @@
MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_power_scalar)
.set_attr("FCompute", BinaryScalarOp::Compute)
.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_power_scalar"});
+MXNET_OPERATOR_REGISTER_NP_BINARY_SCALAR(_npi_logaddexp_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute)
+.set_attr("Fgradient",
ElemwiseGradUseIn{"_backward_logaddexp_scalar"});
+
+
+MXNET_OPERATOR_REGISTER_BINARY(_backward_logaddexp_scalar)
+.add_argument("scalar", "float", "scalar value")
+.set_attr_parser([](NodeAttrs *attrs) { attrs->parsed =
std::stod(attrs->dict["scalar"]); })
+.set_attr("FCompute", BinaryScalarOp::Backward<
+ cpu, mshadow_op::logadd_left>);
+
+
+
+
+
Review comment:
> too many blank lines here
blank lines removed, thanks for reviewing
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] iblis17 commented on a change in pull request #15454: Julia docs
iblis17 commented on a change in pull request #15454: Julia docs
URL: https://github.com/apache/incubator-mxnet/pull/15454#discussion_r302345170
##
File path: docs/install/ubuntu_setup.md
##
@@ -310,25 +310,93 @@ Refer to the [Clojure setup
guide](https://github.com/apache/incubator-mxnet/tre
### Install the MXNet Package for Julia
-The MXNet package for Julia is hosted in a separate repository, MXNet.jl,
which is available on [GitHub](https://github.com/dmlc/MXNet.jl). To use Julia
binding it with an existing libmxnet installation, set the ```MXNET_HOME```
environment variable by running the following command:
+ Install Julia
+The package available through `apt-get` is old and not compatible with the
latest version of MXNet.
+Fetch the latest version (1.0.3 at the time of this writing).
```bash
-export MXNET_HOME=//libmxnet
+wget -qO julia-10.tar.gz
https://julialang-s3.julialang.org/bin/linux/x64/1.0/julia-1.0.3-linux-x86_64.tar.gz
```
-The path to the existing libmxnet installation should be the root directory of
libmxnet. In other words, you should be able to find the ```libmxnet.so``` file
at ```$MXNET_HOME/lib```. For example, if the root directory of libmxnet is
```~```, you would run the following command:
+Place the extracted files somewhere like a julia folder in your home dir.
```bash
-export MXNET_HOME=/~/libmxnet
+mkdir ~/julia
+mv julia-10.tar.gz ~/julia
+cd ~/julia
+tar xvf julia-10.tar.gz
```
-You might want to add this command to your ```~/.bashrc``` file. If you do,
you can install the Julia package in the Julia console using the following
command:
+Test Julia.
+```bash
+cd julia-1.0.3/bin
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+If you're still getting the old version, remove it.
+```bash
+sudo apt remove julia
+```
+
+Update your PATH to have Julia's new location. Add this to your `.zshrc`,
`.bashrc`, `.profile` or `.bash_profile`.
+```bash
+export PATH=~/julia/julia-1.0.3/bin:$PATH
+```
+
+Validate your PATH.
+```bash
+echo $PATH
+```
+
+Validate Julia works and is the expected version.
+```bash
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+ Setup Your MXNet-Julia Environment
+
+**For each of the following environment variables, add the commands to your
`.zshrc`, `.bashrc`, `.profile` or `.bash_profile` to make them persist.**
+
+Create a `julia-depot` folder and environment variable.
+```bash
+mkdir julia-depot
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+```
+
+To use the Julia binding with an existing `libmxnet` installation, set the
`MXNET_HOME` environment variable to the MXNet source root. For example:
+```bash
+export MXNET_HOME=$HOME/incubator-mxnet
+```
-```julia
-Pkg.add("MXNet")
+Now set the `LD_LIBRARY_PATH` environment variable to where `libmxnet.so` is
found. If you can't find it, you might have skipped the building MXNet step. Go
back and [build MXNet](#build-the-shared-library) first. For example:
+```bash
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+```
+
+Verify the location of `libjemalloc.so` and set the `LD_PRELOAD` environment
variable.
+```bash
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+With all of these updates, here's an example of what you might want to have in
your `.zshrc`, `.bashrc`, `.profile` or `.bash_profile`.
+
+```
+export PATH=$HOME/bin:$HOME/.local/bin:$HOME/julia/julia-1.0.3/bin:$PATH
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+export MXNET_HOME=$HOME/incubator-mxnet
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+Install MXNet with Julia:
+
+```bash
+julia --color=yes --project=./ -e \
+ 'using Pkg; \
+ Pkg.develop(PackageSpec(name="MXNet", path =
joinpath(ENV["MXNET_HOME"], "julia")))'
Review comment:
I understand `Pkg.add` do not work at this moment, since v1.5.0 doesn't get
released.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] francis0407 commented on a change in pull request #15495: [Numpy] Added operator logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
francis0407 commented on a change in pull request #15495: [Numpy] Added
operator logaddexp; added support for zero-size tensor in
BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15495#discussion_r302345771
##
File path: python/mxnet/symbol/numpy/_symbol.py
##
@@ -1070,6 +1070,61 @@ def mod(x1, x2, out=None):
def power(x1, x2, out=None):
return _ufunc_helper(x1, x2, _npi.power, _np.power, _npi.power_scalar,
_npi.rpower_scalar, out)
+@set_module('mxnet.symbol.numpy')
+def logaddexp(x1, x2, out=None):
+"""Logarithm of the sum of exponentiations of the inputs.
+logaddexp(x1, x2, out=None)
+
+Calculates ``log(exp(x1) + exp(x2))``. This function is useful in
+statistics where the calculated probabilities of events may be so small
+as to exceed the range of normal floating point numbers. In such cases
+the logarithm of the calculated probability is stored. This function
+allows adding probabilities stored in such a fashion.
+
+Parameters
+--
+x1, x2 : _Symbol or scalar
+Input values.
+out : ndarray, None, or tuple of ndarray and None, optional
+A location into which the result is stored. If provided, it must have
+a shape and dtype as the expected output. If not provided or `None`,
+a freshly-allocated array is returned.
+
+Returns
+---
+result : _Symbol
+Logarithm of ``exp(x1) + exp(x2)``.
+This is a scalar if both `x1` and `x2` are scalars.
+
+See Also
+
+logaddexp2: Logarithm of the sum of exponentiations of inputs in base 2.
+
+Notes
+-
+This function differs from the original `numpy.logaddexp2
+
`_
in
+the following aspects:
+
+- Input type does not support Python native iterables(list, tuple, ...).
Only ndarray is supported.
+- ``out`` param: cannot perform auto broadcasting. ``out`` ndarray's shape
must be the same as the expected output.
+- ``out`` param: cannot perform auto type cast. ``out`` ndarray's dtype
must be the same as the expected output.
+- ``out`` param does not support scalar input case.
+
+Examples
+
+>>> prob1 = np.log(1e-50)
+>>> prob2 = np.log(2.5e-50)
+>>> prob12 = np.logaddexp(prob1, prob2)
+>>> prob12
+-113.87649168120691
+>>> np.exp(prob12)
+3.5057e-50
+"""
+return _ufunc_helper(x1, x2, _npi.logaddexp, _np.logaddexp,
_npi.logaddexp_scalar,
+ _npi.logaddexp_scalar, out)
+
Review comment:
Only two blank lines are needed here.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] francis0407 commented on a change in pull request #15495: [Numpy] Added operator logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
francis0407 commented on a change in pull request #15495: [Numpy] Added
operator logaddexp; added support for zero-size tensor in
BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15495#discussion_r302344674
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cc
##
@@ -150,6 +151,29 @@ Example::
.set_attr("FCompute", BinaryBroadcastCompute)
.set_attr("FGradient",
ElemwiseGradUseIn{"_backward_broadcast_power"});
+MXNET_OPERATOR_REGISTER_BINARY_BROADCAST(_npi_logaddexp)
+.describe(
Review comment:
`.describe("logaddexp" ADD_FILELINE)` would be better
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] francis0407 commented on a change in pull request #15495: [Numpy] Added operator logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
francis0407 commented on a change in pull request #15495: [Numpy] Added operator logaddexp; added support for zero-size tensor in BinaryBroadcastBackwardUseIn URL: https://github.com/apache/incubator-mxnet/pull/15495#discussion_r302345396 ## File path: tests/python/unittest/test_numpy_op.py ## @@ -1284,6 +1284,88 @@ def g(data, axis1, axis2, offset): continue assert False +@with_seed() +@npx.use_np_shape +def test_np_logaddexp(): +@npx.use_np_shape +class TestLogaddexp(HybridBlock): +def __init__(self): +super(TestLogaddexp, self).__init__() + +def hybrid_forward(self, F, x1, x2): +return F.np.logaddexp(x1, x2) + +shapes = [ +((3, 1), (3, 1)), +((3, 1, 2), (3, 1, 2)), +((1, ),(1, )), +((3, 0), (3, 0)), # zero-size shape +((0, 1), (0, 1)), # zero-size shape +((2, 0, 2), (2, 0, 2)), # zero-size shape +((1, ), (3, )), # broadcast +((2, 3), (2, 1)), # broadcast +((1, 3), (2, 3)), # broadcast +((1,3), (2, 0, 3)), # broadcast to zero-dim shape +((1, 0, 1), (3, 0, 1)), # broadcast of zero-dim shape +((), ()), # zero-dim shape +] +eps = 1e-3 +# Legal shape test. +for shape_a, shape_b in shapes: +for hybridize in [True, False]: +test_logaddexp = TestLogaddexp() +if hybridize: +test_logaddexp.hybridize() +lhs = rand_ndarray(shape_a).as_np_ndarray() +rhs = rand_ndarray(shape_b).as_np_ndarray() +lhs.attach_grad() +rhs.attach_grad() +np_out = _np.logaddexp(lhs.asnumpy(), rhs.asnumpy()) +np_backward_lhs = _np.exp(lhs.asnumpy()) / (_np.exp(lhs.asnumpy()) + _np.exp(rhs.asnumpy())) +np_backward_rhs = _np.exp(rhs.asnumpy()) / (_np.exp(lhs.asnumpy()) + _np.exp(rhs.asnumpy())) +with mx.autograd.record(): +mx_out = test_logaddexp(lhs, rhs) +assert mx_out.shape == np_out.shape +assert_almost_equal(mx_out.asnumpy(), np_out, rtol=1e-3, atol=1e-5) +mx_out.backward() +# For broadcast backward case, +# reduce sum is applied on numpy result. +for n_dim in range(len(shape_a)): +if (shape_a[n_dim] != shape_b[n_dim]): +if (shape_a[n_dim] > shape_b[n_dim]): +np_backward_rhs = np_backward_rhs.sum(axis=n_dim, keepdims=True) +else: +np_backward_lhs = np_backward_lhs.sum(axis=n_dim, keepdims=True) +assert_almost_equal(lhs.grad.asnumpy(), np_backward_lhs, rtol=1e-3, atol=1e-5) +assert_almost_equal(rhs.grad.asnumpy(), np_backward_rhs, rtol=1e-3, atol=1e-5) +# Test imperative once again +mx_out = np.logaddexp(lhs, rhs) +np_out = _np.logaddexp(lhs.asnumpy(), rhs.asnumpy()) +assert_almost_equal(mx_out.asnumpy(), np_out, rtol=1e-3, atol=1e-5) + +# Range case. +x = [100, -100, 1000200, -1000200] +y = [1000200, -1000200, 100, -100] +z = [1000200, -100, 1000200, -100] +for dt in ['float64']: +logxf = np.array(x, dtype=dt) +logyf = np.array(y, dtype=dt) +logzf = np.array(z, dtype=dt) +assert_almost_equal(np.logaddexp(logxf, logyf).asnumpy(), +logzf.asnumpy()) + + Review comment: One blank line is enough here. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] iblis17 commented on a change in pull request #15454: Julia docs
iblis17 commented on a change in pull request #15454: Julia docs
URL: https://github.com/apache/incubator-mxnet/pull/15454#discussion_r302346283
##
File path: docs/install/ubuntu_setup.md
##
@@ -310,25 +310,93 @@ Refer to the [Clojure setup
guide](https://github.com/apache/incubator-mxnet/tre
### Install the MXNet Package for Julia
-The MXNet package for Julia is hosted in a separate repository, MXNet.jl,
which is available on [GitHub](https://github.com/dmlc/MXNet.jl). To use Julia
binding it with an existing libmxnet installation, set the ```MXNET_HOME```
environment variable by running the following command:
+ Install Julia
+The package available through `apt-get` is old and not compatible with the
latest version of MXNet.
+Fetch the latest version (1.0.3 at the time of this writing).
```bash
-export MXNET_HOME=//libmxnet
+wget -qO julia-10.tar.gz
https://julialang-s3.julialang.org/bin/linux/x64/1.0/julia-1.0.3-linux-x86_64.tar.gz
```
-The path to the existing libmxnet installation should be the root directory of
libmxnet. In other words, you should be able to find the ```libmxnet.so``` file
at ```$MXNET_HOME/lib```. For example, if the root directory of libmxnet is
```~```, you would run the following command:
+Place the extracted files somewhere like a julia folder in your home dir.
```bash
-export MXNET_HOME=/~/libmxnet
+mkdir ~/julia
+mv julia-10.tar.gz ~/julia
+cd ~/julia
+tar xvf julia-10.tar.gz
```
-You might want to add this command to your ```~/.bashrc``` file. If you do,
you can install the Julia package in the Julia console using the following
command:
+Test Julia.
+```bash
+cd julia-1.0.3/bin
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+If you're still getting the old version, remove it.
+```bash
+sudo apt remove julia
+```
+
+Update your PATH to have Julia's new location. Add this to your `.zshrc`,
`.bashrc`, `.profile` or `.bash_profile`.
+```bash
+export PATH=~/julia/julia-1.0.3/bin:$PATH
+```
+
+Validate your PATH.
+```bash
+echo $PATH
+```
+
+Validate Julia works and is the expected version.
+```bash
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+ Setup Your MXNet-Julia Environment
+
+**For each of the following environment variables, add the commands to your
`.zshrc`, `.bashrc`, `.profile` or `.bash_profile` to make them persist.**
+
+Create a `julia-depot` folder and environment variable.
+```bash
+mkdir julia-depot
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+```
+
+To use the Julia binding with an existing `libmxnet` installation, set the
`MXNET_HOME` environment variable to the MXNet source root. For example:
+```bash
+export MXNET_HOME=$HOME/incubator-mxnet
+```
-```julia
-Pkg.add("MXNet")
+Now set the `LD_LIBRARY_PATH` environment variable to where `libmxnet.so` is
found. If you can't find it, you might have skipped the building MXNet step. Go
back and [build MXNet](#build-the-shared-library) first. For example:
+```bash
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+```
+
+Verify the location of `libjemalloc.so` and set the `LD_PRELOAD` environment
variable.
+```bash
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+With all of these updates, here's an example of what you might want to have in
your `.zshrc`, `.bashrc`, `.profile` or `.bash_profile`.
+
+```
+export PATH=$HOME/bin:$HOME/.local/bin:$HOME/julia/julia-1.0.3/bin:$PATH
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+export MXNET_HOME=$HOME/incubator-mxnet
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+Install MXNet with Julia:
+
+```bash
+julia --color=yes --project=./ -e \
+ 'using Pkg; \
+ Pkg.develop(PackageSpec(name="MXNet", path =
joinpath(ENV["MXNET_HOME"], "julia")))'
Review comment:
Can we still change the web page via PR after a release rolled out, or the
web page only update once for a single release?
If yes, I'm okay for waiting the package registry setup.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on a change in pull request #15497: Independent gradients requests check with respect to weights and bias of convolution
pengzhao-intel commented on a change in pull request #15497: Independent
gradients requests check with respect to weights and bias of convolution
URL: https://github.com/apache/incubator-mxnet/pull/15497#discussion_r302346944
##
File path: src/operator/nn/mkldnn/mkldnn_convolution.cc
##
@@ -662,21 +662,21 @@ void MKLDNNConvolutionBackward(const nnvm::NodeAttrs&
attrs, const OpContext &ct
in_grad[conv::kWeight],
convBwdWeight.bwdWeights_pd.diff_weights_primitive_desc(),
req[conv::kWeight]);
-mkldnn_output_t in_grad_bias;
-if (param.no_bias) {
- convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second);
- MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
-} else {
- in_grad_bias = CreateMKLDNNMem(
+
+if (!param.no_bias && req[conv::kBias]) {
+ auto in_grad_bias = CreateMKLDNNMem(
in_grad[conv::kBias],
convBwdWeight.bwdWeights_pd.diff_bias_primitive_desc(),
req[conv::kBias]);
convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second, *in_grad_bias.second);
+ *in_grad_weight.second, *in_grad_bias.second);
MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
CommitOutput(in_grad[conv::kBias], in_grad_bias);
+} else if (req[conv::kWeight]) {
+ convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
+ *in_grad_weight.second);
+ MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
}
-CommitOutput(in_grad[conv::kWeight], in_grad_weight);
+if (req[conv::kWeight]) CommitOutput(in_grad[conv::kWeight],
in_grad_weight);
Review comment:
what's the behavior of req[conv::bias]?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #15448: [MKLDNN]Enhance Quantization APIs and Tutorial
pengzhao-intel commented on issue #15448: [MKLDNN]Enhance Quantization APIs and Tutorial URL: https://github.com/apache/incubator-mxnet/pull/15448#issuecomment-510310011 @ThomasDelteil Would you mind to take a review again? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #15130: Add NaiveEngine tests in CI
pengzhao-intel commented on issue #15130: Add NaiveEngine tests in CI URL: https://github.com/apache/incubator-mxnet/pull/15130#issuecomment-510311947 Closing this PR. We're not urgent to enable naive engine test. Let's wait if someone requests this in the future. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302348771 ## File path: python/mxnet/ndarray/numpy/_op.py ## @@ -33,7 +33,88 @@ 'clip', 'split', 'swapaxes', 'expand_dims', 'tile', 'linspace', 'sin', 'cos', 'sinh', 'cosh', 'log10', 'sqrt', 'abs', 'exp', 'arctan', 'sign', 'log', 'degrees', 'log2', 'rint', 'radians', 'mean', 'reciprocal', 'square', 'arcsin', - 'argsort'] + 'argsort', 'tensordot'] + Review comment: Fixed. Thx. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel closed pull request #15130: Add NaiveEngine tests in CI
pengzhao-intel closed pull request #15130: Add NaiveEngine tests in CI URL: https://github.com/apache/incubator-mxnet/pull/15130 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302348845 ## File path: python/mxnet/numpy/multiarray.py ## @@ -45,10 +45,90 @@ __all__ = ['ndarray', 'empty', 'array', 'zeros', 'ones', 'maximum', 'minimum', 'stack', 'arange', 'argmax', 'add', 'subtract', 'multiply', 'divide', 'mod', 'power', 'concatenate', - 'clip', 'split', 'swapaxes', 'expand_dims', 'tile', 'linspace', 'sin', 'cos', 'sin', 'cos', 'sinh', 'cosh', 'log10', 'sqrt', 'abs', 'exp', 'arctan', 'sign', 'log', 'degrees', 'log2', 'rint', 'radians', 'mean', 'reciprocal', 'square', 'arcsin', - 'argsort'] + 'argsort', 'tensordot'] + Review comment: Fixed. Thx. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] anirudh2290 merged pull request #15455: Improve docs for AMP
anirudh2290 merged pull request #15455: Improve docs for AMP URL: https://github.com/apache/incubator-mxnet/pull/15455 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on issue #15349: Numpy Tensordot Operator
reminisce commented on issue #15349: Numpy Tensordot Operator URL: https://github.com/apache/incubator-mxnet/pull/15349#issuecomment-510312746 Please fix the CI failures. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302349667
##
File path: tests/python/unittest/test_numpy_op.py
##
@@ -26,7 +26,151 @@
from mxnet.test_utils import check_numeric_gradient
from common import assertRaises, with_seed
import random
+import collections
+@with_seed()
+@npx.use_np_shape
+def test_np_tensordot():
+class TestTensordot(HybridBlock):
+def __init__(self, axes):
+super(TestTensordot, self).__init__()
+self._axes = axes
+
+def hybrid_forward(self, F, a, b):
+return F.np.tensordot(a, b, self._axes)
+
+def tensordot_backward(a, b, axes = 2):
+if (a.ndim < 1) or (b.ndim < 1):
+raise ValueError('An input is zero-dim')
+
+if isinstance(axes, collections.abc.Sequence):
+if len(axes) != 2:
+raise ValueError('Axes must consist of two arrays.')
+a_axes_summed, b_axes_summed = axes
+if _np.isscalar(a_axes_summed):
+a_axes_summed = a_axes_summed,
+if _np.isscalar(b_axes_summed):
+b_axes_summed = b_axes_summed,
+else:
+a_axes_summed = [i + a.ndim - axes for i in range(axes)]
+b_axes_summed = [i for i in range(axes)]
+
+if len(a_axes_summed) != len(b_axes_summed):
+raise ValueError('Axes length mismatch')
+
+a_axes_remained = []
+for i in range(a.ndim):
+if not (i in a_axes_summed):
+a_axes_remained.append(i)
+a_axes = a_axes_remained[:] + a_axes_summed[:]
+
+b_axes_remained = []
+for i in range(b.ndim):
+if not (i in b_axes_summed):
+b_axes_remained.append(i)
+b_axes = b_axes_summed[:] + b_axes_remained[:]
+
+ad1 = _np.prod([a.shape[i] for i in a_axes_remained]) if
len(a_axes_remained) > 0 else 1
+ad2 = _np.prod([a.shape[i] for i in a_axes_summed]) if
len(a_axes_summed) > 0 else 1
+bd1 = _np.prod([b.shape[i] for i in b_axes_summed]) if
len(b_axes_summed) > 0 else 1
+bd2 = _np.prod([b.shape[i] for i in b_axes_remained]) if
len(b_axes_remained) > 0 else 1
+
+out_grad = _np.ones((ad1, bd2))
+
+new_a = _np.transpose(a, a_axes)
+new_a_shape = new_a.shape[:]
+new_a = new_a.reshape((ad1, ad2))
+new_b = _np.transpose(b, b_axes)
+new_b_shape = new_b.shape[:]
+new_b = new_b.reshape((bd1, bd2))
+
+reverse_a_axes = [0 for i in a_axes]
+for i in range(len(a_axes)):
+reverse_a_axes[a_axes[i]] = i
+
+reverse_b_axes = [0 for i in b_axes]
+for i in range(len(b_axes)):
+reverse_b_axes[b_axes[i]] = i
+
+grad_b = _np.dot(new_a.T, out_grad).reshape(new_b_shape)
+grad_b = _np.transpose(grad_b, reverse_b_axes)
+grad_a = _np.dot(out_grad, new_b.T).reshape(new_a_shape)
+grad_a = _np.transpose(grad_a, reverse_a_axes)
+
+return [grad_a, grad_b]
+
+# test non zero size input
+tensor_shapes = [
+((3, 5), (5, 4), 1), # (a_shape, b_shape, axes)
+((3,), (3,), 1),
+((3, 4, 5, 6, 7), (5, 6, 7, 1, 2), 3),
+((3, 5, 4, 6, 7), (7, 6, 5, 1, 2), [[1, 3, 4], [2, 1, 0]]),
+((2, 2), (2, 2), 2),
+((3, 5, 4), (5, ), [[1], [0]]),
+((2,), (2, 3), 1),
+((3,), (3,), 0),
+((2,), (2, 3), 0),
+((3, 5, 4), (5, ), 0)
+]
+
+for hybridize in [True, False]:
+for a_shape, b_shape, axes in tensor_shapes:
+for dtype in [_np.float32, _np.float64]:
+test_tensordot = TestTensordot(axes)
+if hybridize:
+test_tensordot.hybridize()
+a = rand_ndarray(shape = a_shape, dtype =
dtype).as_np_ndarray()
+b = rand_ndarray(shape = b_shape, dtype =
dtype).as_np_ndarray()
+a.attach_grad()
+b.attach_grad()
+
+np_out = _np.tensordot(a.asnumpy(), b.asnumpy(), axes)
+with mx.autograd.record():
+mx_out = test_tensordot(a, b)
+assert mx_out.shape == np_out.shape
+assert_almost_equal(mx_out.asnumpy(), np_out, rtol = 1e-3,
atol = 1e-5)
+mx_out.backward()
+np_backward = tensordot_backward(a.asnumpy(), b.asnumpy(),
axes)
+assert_almost_equal(a.grad.asnumpy(), np_backward[0], rtol =
1e-3, atol=1e-5)
+assert_almost_equal(b.grad.asnumpy(), np_backward[1], rtol =
1e-3, atol=1e-5)
+
+# Test imperative once again
+mx_out = np.tensordot(a, b, axes)
+np_out = _np.tensordot(a.asnumpy(), b.asnumpy(), axes)
+assert_almo
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302349487
##
File path: src/operator/numpy/np_tensordot_op-inl.h
##
@@ -0,0 +1,399 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_tensordot_op-inl.h
+ * \brief CPU Implementation of numpy-compatible tensordot
+ */
+#ifndef MXNET_OPERATOR_NUMPY_NP_TENSORDOT_OP_INL_H_
+#define MXNET_OPERATOR_NUMPY_NP_TENSORDOT_OP_INL_H_
+
+#include
+#include "np_matrix_op-inl.h"
+
+namespace mxnet {
+namespace op {
+
+using namespace mxnet;
+using namespace mshadow;
+
+struct TensordotParam : public dmlc::Parameter {
+ mxnet::Tuple a_axes_summed, b_axes_summed;
+ DMLC_DECLARE_PARAMETER(TensordotParam) {
+DMLC_DECLARE_FIELD(a_axes_summed);
+DMLC_DECLARE_FIELD(b_axes_summed);
+ }
+};
+
+/**
+ * Gets matrix dimensions of a and b after transpose and reshape.
+ */
+inline void GetMatrixDimensions(
+int* ad1,
+int* ad2,
+int* bd1,
+int* bd2,
Review comment:
We need to change these values in the function. Using reference can't pass
the sanity check.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302350329 ## File path: src/operator/numpy/np_tensordot_int_axes_op-inl.h ## @@ -0,0 +1,213 @@ +/* Review comment: Fixed. Thx. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel commented on issue #15491: [Doc] Add MKL install method apt/yum into tutorial
pengzhao-intel commented on issue #15491: [Doc] Add MKL install method apt/yum into tutorial URL: https://github.com/apache/incubator-mxnet/pull/15491#issuecomment-510314614 Thanks, merging now :) This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] pengzhao-intel merged pull request #15491: [Doc] Add MKL install method apt/yum into tutorial
pengzhao-intel merged pull request #15491: [Doc] Add MKL install method apt/yum into tutorial URL: https://github.com/apache/incubator-mxnet/pull/15491 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302351664
##
File path: src/operator/numpy/np_tensordot_int_axes_op-inl.h
##
@@ -0,0 +1,213 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_tensordot_int_axes_op-inl.h
+ * \brief Implementation of numpy-compatible tensordot_int_axes
+ */
+#ifndef MXNET_OPERATOR_NUMPY_NP_TENSORDOT_INT_AXES_OP_INL_H_
+#define MXNET_OPERATOR_NUMPY_NP_TENSORDOT_INT_AXES_OP_INL_H_
+
+#include
+#include "np_tensordot_op-inl.h"
+
+namespace mxnet {
+namespace op {
+
+using namespace mxnet;
+using namespace mshadow;
+
+struct TensordotIntAxesParam : public dmlc::Parameter {
+ int axes;
+ DMLC_DECLARE_PARAMETER(TensordotIntAxesParam) {
+DMLC_DECLARE_FIELD(axes);
+ }
+};
+
+/**
+ * gets summed axes of a and b from parameter axes.
+ */
+inline void GetSummedAxes(
+mxnet::Tuple* a_axes_summed_ptr,
+mxnet::Tuple* b_axes_summed_ptr,
+const int axes,
+const mxnet::TShape& a_shape) {
+ std::vector a_axes_summed_vector;
+ for (int i = 0; i < axes; i++) {
+a_axes_summed_vector.push_back(a_shape.ndim() - axes + i);
+ }
+ *a_axes_summed_ptr = mxnet::Tuple(a_axes_summed_vector);
+
+ std::vector b_axes_summed_vector;
+ for (int i = 0; i < axes; i++) {
+b_axes_summed_vector.push_back(i);
+ }
+ *b_axes_summed_ptr = mxnet::Tuple(b_axes_summed_vector);
+}
+
+/**
+ * Calculates tensordot.
+ */
+template
+void TensordotIntAxesImpl(
Review comment:
Fixed. Thx.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302351820
##
File path: src/operator/numpy/np_tensordot_int_axes_op-inl.h
##
@@ -0,0 +1,213 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_tensordot_int_axes_op-inl.h
+ * \brief Implementation of numpy-compatible tensordot_int_axes
+ */
+#ifndef MXNET_OPERATOR_NUMPY_NP_TENSORDOT_INT_AXES_OP_INL_H_
+#define MXNET_OPERATOR_NUMPY_NP_TENSORDOT_INT_AXES_OP_INL_H_
+
+#include
+#include "np_tensordot_op-inl.h"
+
+namespace mxnet {
+namespace op {
+
+using namespace mxnet;
+using namespace mshadow;
+
+struct TensordotIntAxesParam : public dmlc::Parameter {
+ int axes;
+ DMLC_DECLARE_PARAMETER(TensordotIntAxesParam) {
+DMLC_DECLARE_FIELD(axes);
+ }
+};
+
+/**
+ * gets summed axes of a and b from parameter axes.
+ */
+inline void GetSummedAxes(
+mxnet::Tuple* a_axes_summed_ptr,
+mxnet::Tuple* b_axes_summed_ptr,
+const int axes,
+const mxnet::TShape& a_shape) {
+ std::vector a_axes_summed_vector;
+ for (int i = 0; i < axes; i++) {
+a_axes_summed_vector.push_back(a_shape.ndim() - axes + i);
+ }
+ *a_axes_summed_ptr = mxnet::Tuple(a_axes_summed_vector);
+
+ std::vector b_axes_summed_vector;
+ for (int i = 0; i < axes; i++) {
+b_axes_summed_vector.push_back(i);
+ }
+ *b_axes_summed_ptr = mxnet::Tuple(b_axes_summed_vector);
+}
+
+/**
+ * Calculates tensordot.
+ */
+template
+void TensordotIntAxesImpl(
+const int axes,
+const OpContext& ctx,
+const TBlob& a,
+const TBlob& b,
+const TBlob& out,
+const std::vector& req) {
+
+ if (req[0] == kNullOp) {
+return;
+ }
+
+ if (out.shape_.Size() == 0U) {
+return; // zero-size output, no need to launch kernel
+ }
+
+ const mxnet::TShape& a_shape = a.shape_;
+ const mxnet::TShape& b_shape = b.shape_;
+
+ mshadow::Stream *s = ctx.get_stream();
+ CHECK_EQ(out.type_flag_, a.type_flag_)
+ << "Binary function only support input/output with the same type";
+ CHECK_EQ(out.type_flag_, b.type_flag_)
+ << "Binary function only support input/output with the same type";
+ CHECK(out.type_flag_ == kFloat32 || out.type_flag_ == kFloat64 ||
Review comment:
Fixed. Thx.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] iblis17 commented on a change in pull request #15454: Julia docs
iblis17 commented on a change in pull request #15454: Julia docs
URL: https://github.com/apache/incubator-mxnet/pull/15454#discussion_r302346283
##
File path: docs/install/ubuntu_setup.md
##
@@ -310,25 +310,93 @@ Refer to the [Clojure setup
guide](https://github.com/apache/incubator-mxnet/tre
### Install the MXNet Package for Julia
-The MXNet package for Julia is hosted in a separate repository, MXNet.jl,
which is available on [GitHub](https://github.com/dmlc/MXNet.jl). To use Julia
binding it with an existing libmxnet installation, set the ```MXNET_HOME```
environment variable by running the following command:
+ Install Julia
+The package available through `apt-get` is old and not compatible with the
latest version of MXNet.
+Fetch the latest version (1.0.3 at the time of this writing).
```bash
-export MXNET_HOME=//libmxnet
+wget -qO julia-10.tar.gz
https://julialang-s3.julialang.org/bin/linux/x64/1.0/julia-1.0.3-linux-x86_64.tar.gz
```
-The path to the existing libmxnet installation should be the root directory of
libmxnet. In other words, you should be able to find the ```libmxnet.so``` file
at ```$MXNET_HOME/lib```. For example, if the root directory of libmxnet is
```~```, you would run the following command:
+Place the extracted files somewhere like a julia folder in your home dir.
```bash
-export MXNET_HOME=/~/libmxnet
+mkdir ~/julia
+mv julia-10.tar.gz ~/julia
+cd ~/julia
+tar xvf julia-10.tar.gz
```
-You might want to add this command to your ```~/.bashrc``` file. If you do,
you can install the Julia package in the Julia console using the following
command:
+Test Julia.
+```bash
+cd julia-1.0.3/bin
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+If you're still getting the old version, remove it.
+```bash
+sudo apt remove julia
+```
+
+Update your PATH to have Julia's new location. Add this to your `.zshrc`,
`.bashrc`, `.profile` or `.bash_profile`.
+```bash
+export PATH=~/julia/julia-1.0.3/bin:$PATH
+```
+
+Validate your PATH.
+```bash
+echo $PATH
+```
+
+Validate Julia works and is the expected version.
+```bash
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+ Setup Your MXNet-Julia Environment
+
+**For each of the following environment variables, add the commands to your
`.zshrc`, `.bashrc`, `.profile` or `.bash_profile` to make them persist.**
+
+Create a `julia-depot` folder and environment variable.
+```bash
+mkdir julia-depot
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+```
+
+To use the Julia binding with an existing `libmxnet` installation, set the
`MXNET_HOME` environment variable to the MXNet source root. For example:
+```bash
+export MXNET_HOME=$HOME/incubator-mxnet
+```
-```julia
-Pkg.add("MXNet")
+Now set the `LD_LIBRARY_PATH` environment variable to where `libmxnet.so` is
found. If you can't find it, you might have skipped the building MXNet step. Go
back and [build MXNet](#build-the-shared-library) first. For example:
+```bash
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+```
+
+Verify the location of `libjemalloc.so` and set the `LD_PRELOAD` environment
variable.
+```bash
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+With all of these updates, here's an example of what you might want to have in
your `.zshrc`, `.bashrc`, `.profile` or `.bash_profile`.
+
+```
+export PATH=$HOME/bin:$HOME/.local/bin:$HOME/julia/julia-1.0.3/bin:$PATH
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+export MXNET_HOME=$HOME/incubator-mxnet
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+Install MXNet with Julia:
+
+```bash
+julia --color=yes --project=./ -e \
+ 'using Pkg; \
+ Pkg.develop(PackageSpec(name="MXNet", path =
joinpath(ENV["MXNET_HOME"], "julia")))'
Review comment:
Can we still change the web page via PR after a release rolled out, or the
web page only update once for a single release?
If yes, I'm okay for waiting the package registry setup, then add some
descriptions for `Pkg.add`.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302353380
##
File path: src/operator/numpy/np_tensordot_int_axes_op-inl.h
##
@@ -0,0 +1,213 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_tensordot_int_axes_op-inl.h
+ * \brief Implementation of numpy-compatible tensordot_int_axes
+ */
+#ifndef MXNET_OPERATOR_NUMPY_NP_TENSORDOT_INT_AXES_OP_INL_H_
+#define MXNET_OPERATOR_NUMPY_NP_TENSORDOT_INT_AXES_OP_INL_H_
+
+#include
+#include "np_tensordot_op-inl.h"
+
+namespace mxnet {
+namespace op {
+
+using namespace mxnet;
+using namespace mshadow;
+
+struct TensordotIntAxesParam : public dmlc::Parameter {
+ int axes;
+ DMLC_DECLARE_PARAMETER(TensordotIntAxesParam) {
+DMLC_DECLARE_FIELD(axes);
+ }
+};
+
+/**
+ * gets summed axes of a and b from parameter axes.
+ */
+inline void GetSummedAxes(
+mxnet::Tuple* a_axes_summed_ptr,
+mxnet::Tuple* b_axes_summed_ptr,
+const int axes,
+const mxnet::TShape& a_shape) {
+ std::vector a_axes_summed_vector;
+ for (int i = 0; i < axes; i++) {
+a_axes_summed_vector.push_back(a_shape.ndim() - axes + i);
+ }
+ *a_axes_summed_ptr = mxnet::Tuple(a_axes_summed_vector);
+
+ std::vector b_axes_summed_vector;
+ for (int i = 0; i < axes; i++) {
+b_axes_summed_vector.push_back(i);
+ }
+ *b_axes_summed_ptr = mxnet::Tuple(b_axes_summed_vector);
+}
+
+/**
+ * Calculates tensordot.
+ */
+template
+void TensordotIntAxesImpl(
+const int axes,
+const OpContext& ctx,
+const TBlob& a,
+const TBlob& b,
+const TBlob& out,
+const std::vector& req) {
+
+ if (req[0] == kNullOp) {
+return;
+ }
+
+ if (out.shape_.Size() == 0U) {
+return; // zero-size output, no need to launch kernel
+ }
+
+ const mxnet::TShape& a_shape = a.shape_;
+ const mxnet::TShape& b_shape = b.shape_;
+
+ mshadow::Stream *s = ctx.get_stream();
+ CHECK_EQ(out.type_flag_, a.type_flag_)
+ << "Binary function only support input/output with the same type";
+ CHECK_EQ(out.type_flag_, b.type_flag_)
+ << "Binary function only support input/output with the same type";
+ CHECK(out.type_flag_ == kFloat32 || out.type_flag_ == kFloat64 ||
+ (out.type_flag_ == kFloat16 && ctx.run_ctx.ctx.dev_mask() ==
mshadow::gpu::kDevMask))
+ << "Tensordot only supports float32/float64 for CPU, and
float16/float32/float64 for GPU";
+
+ Tuple a_axes_summed;
+ Tuple b_axes_summed;
+ GetSummedAxes(&a_axes_summed, &b_axes_summed, axes, a_shape);
+
+ Tuple a_axes_remained;
+ Tuple b_axes_remained;
+ Tuple a_axes;
+ Tuple b_axes;
+ GetReorderedAxes(a_axes_summed, &a_axes_remained, &a_axes, b_axes_summed,
&b_axes_remained,
+&b_axes, a_shape, b_shape);
Review comment:
Fixed. Thx.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302353567
##
File path: src/operator/numpy/np_tensordot_op.cc
##
@@ -0,0 +1,186 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_tensordot_op.cc
+ * \brief CPU Implementation of numpy-compatible tensordot
+ */
+
+#include
+#include "np_tensordot_op-inl.h"
+
+namespace mxnet {
+namespace op {
+
+bool TensordotOpShape(const nnvm::NodeAttrs& attrs,
+ mxnet::ShapeVector *in_attrs,
Review comment:
Fixed. Thx.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] endvroy commented on a change in pull request #15293: Numpy bitwise OR operator
endvroy commented on a change in pull request #15293: Numpy bitwise OR operator
URL: https://github.com/apache/incubator-mxnet/pull/15293#discussion_r302354149
##
File path: src/operator/numpy/np_elemwise_binary_op.cc
##
@@ -0,0 +1,51 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_elemwise_binary_op.cc
+ * \brief CPU Implementation of numpy-compatible element-wise binary operations
+ */
+
+#include
+#include "../mshadow_op.h"
+#include "../operator_common.h"
+#include "../tensor/elemwise_binary_op.h"
+#include "../tensor/elemwise_binary_broadcast_op.h"
+
+namespace mxnet {
+namespace op {
+
+NNVM_REGISTER_OP(_np_bitwise_or)
+.set_num_inputs(2)
+.set_num_outputs(1)
+.set_attr("FInferShape", BinaryBroadcastShape)
+.set_attr("FInferType", ElemwiseType<2, 1>)
Review comment:
fixed in the most recent commit
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] iblis17 commented on a change in pull request #15454: Julia docs
iblis17 commented on a change in pull request #15454: Julia docs
URL: https://github.com/apache/incubator-mxnet/pull/15454#discussion_r302354265
##
File path: docs/install/ubuntu_setup.md
##
@@ -310,25 +310,93 @@ Refer to the [Clojure setup
guide](https://github.com/apache/incubator-mxnet/tre
### Install the MXNet Package for Julia
-The MXNet package for Julia is hosted in a separate repository, MXNet.jl,
which is available on [GitHub](https://github.com/dmlc/MXNet.jl). To use Julia
binding it with an existing libmxnet installation, set the ```MXNET_HOME```
environment variable by running the following command:
+ Install Julia
+The package available through `apt-get` is old and not compatible with the
latest version of MXNet.
+Fetch the latest version (1.0.3 at the time of this writing).
```bash
-export MXNET_HOME=//libmxnet
+wget -qO julia-10.tar.gz
https://julialang-s3.julialang.org/bin/linux/x64/1.0/julia-1.0.3-linux-x86_64.tar.gz
```
-The path to the existing libmxnet installation should be the root directory of
libmxnet. In other words, you should be able to find the ```libmxnet.so``` file
at ```$MXNET_HOME/lib```. For example, if the root directory of libmxnet is
```~```, you would run the following command:
+Place the extracted files somewhere like a julia folder in your home dir.
```bash
-export MXNET_HOME=/~/libmxnet
+mkdir ~/julia
+mv julia-10.tar.gz ~/julia
+cd ~/julia
+tar xvf julia-10.tar.gz
```
-You might want to add this command to your ```~/.bashrc``` file. If you do,
you can install the Julia package in the Julia console using the following
command:
+Test Julia.
+```bash
+cd julia-1.0.3/bin
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+If you're still getting the old version, remove it.
+```bash
+sudo apt remove julia
+```
+
+Update your PATH to have Julia's new location. Add this to your `.zshrc`,
`.bashrc`, `.profile` or `.bash_profile`.
+```bash
+export PATH=~/julia/julia-1.0.3/bin:$PATH
+```
+
+Validate your PATH.
+```bash
+echo $PATH
+```
+
+Validate Julia works and is the expected version.
+```bash
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+ Setup Your MXNet-Julia Environment
+
+**For each of the following environment variables, add the commands to your
`.zshrc`, `.bashrc`, `.profile` or `.bash_profile` to make them persist.**
+
+Create a `julia-depot` folder and environment variable.
+```bash
+mkdir julia-depot
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+```
+
+To use the Julia binding with an existing `libmxnet` installation, set the
`MXNET_HOME` environment variable to the MXNet source root. For example:
+```bash
+export MXNET_HOME=$HOME/incubator-mxnet
+```
-```julia
-Pkg.add("MXNet")
+Now set the `LD_LIBRARY_PATH` environment variable to where `libmxnet.so` is
found. If you can't find it, you might have skipped the building MXNet step. Go
back and [build MXNet](#build-the-shared-library) first. For example:
+```bash
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+```
+
+Verify the location of `libjemalloc.so` and set the `LD_PRELOAD` environment
variable.
+```bash
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+With all of these updates, here's an example of what you might want to have in
your `.zshrc`, `.bashrc`, `.profile` or `.bash_profile`.
+
+```
+export PATH=$HOME/bin:$HOME/.local/bin:$HOME/julia/julia-1.0.3/bin:$PATH
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
Review comment:
> This combines with Pkg.develop?
This is not necessary for `Pkg.develop`. Actually, there is a default value
for `JULIA_DEPOT_PATH`, and the default is fine for most of user.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
ckt624 commented on a change in pull request #15349: Numpy Tensordot Operator
URL: https://github.com/apache/incubator-mxnet/pull/15349#discussion_r302354817
##
File path: src/operator/numpy/np_tensordot_op.cc
##
@@ -0,0 +1,186 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_tensordot_op.cc
+ * \brief CPU Implementation of numpy-compatible tensordot
+ */
+
+#include
+#include "np_tensordot_op-inl.h"
+
+namespace mxnet {
+namespace op {
+
+bool TensordotOpShape(const nnvm::NodeAttrs& attrs,
+ mxnet::ShapeVector *in_attrs,
+ mxnet::ShapeVector *out_attrs) {
+ CHECK_EQ(in_attrs->size(), 2U);
+ CHECK_EQ(out_attrs->size(), 1U);
+
+ const mxnet::TShape& a_shape = in_attrs->at(0);
+ const mxnet::TShape& b_shape = in_attrs->at(1);
+
+ if (!ndim_is_known(a_shape) || !ndim_is_known(b_shape)) {
+return false;
+ }
+
+ if ((a_shape.ndim() < 1) || (b_shape.ndim() < 1)) {
+return false;
+ }
+
+ const TensordotParam& param = nnvm::get(attrs.parsed);
+ const Tuple& a_axes_summed = param.a_axes_summed;
+ const Tuple& b_axes_summed = param.b_axes_summed;
+
+ Tuple a_axes_remained;
+ Tuple b_axes_remained;
+ Tuple a_axes;
+ Tuple b_axes;
+ GetReorderedAxes(a_axes_summed, &a_axes_remained, &a_axes, b_axes_summed,
&b_axes_remained,
+&b_axes, a_shape, b_shape);
+
+ CHECK_EQ(a_axes_summed.ndim(), b_axes_summed.ndim());
+
+ mxnet::TShape out_shape(a_axes_remained.ndim() + b_axes_remained.ndim(), -1);
+ for (int i = 0; i < a_axes_remained.ndim(); i++) {
+out_shape[i] = a_shape[a_axes_remained[i]];
+ }
+ for (int i = 0; i < b_axes_remained.ndim(); i++) {
+out_shape[a_axes_remained.ndim() + i] = b_shape[b_axes_remained[i]];
+ }
+ SHAPE_ASSIGN_CHECK(*out_attrs, 0, out_shape);
+
+ mxnet::TShape tem_shape1(a_axes.ndim(), -1);
+ for (int i = 0; i < a_axes_remained.ndim(); i++) {
+tem_shape1[a_axes_remained[i]] = out_shape[i];
+ }
+ for (int i = 0; i < a_axes_summed.ndim(); i++) {
+tem_shape1[a_axes_summed[i]] = b_shape[b_axes_summed[i]];
+ }
+ SHAPE_ASSIGN_CHECK(*in_attrs, 0, tem_shape1);
+
+ mxnet::TShape tem_shape2(b_axes.ndim(), -1);
+ for (int i = 0; i < b_axes_remained.ndim(); i++) {
+tem_shape2[b_axes_remained[i]] = out_shape[a_axes_remained.ndim() + i];
+ }
+ for (int i = 0; i < b_axes_summed.ndim(); i++) {
+tem_shape2[b_axes_summed[i]] = a_shape[a_axes_summed[i]];
+ }
+ SHAPE_ASSIGN_CHECK(*in_attrs, 1, tem_shape2);
+
+ return shape_is_known(*in_attrs) && shape_is_known(*out_attrs);
+}
+
+DMLC_REGISTER_PARAMETER(TensordotParam);
+
+NNVM_REGISTER_OP(tensordot)
Review comment:
Fixed. Thx.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on a change in pull request #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
reminisce commented on a change in pull request #15494: [Numpy] Added operator
logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15494#discussion_r302350415
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cu
##
@@ -48,6 +48,14 @@ NNVM_REGISTER_OP(_npi_maximum)
NNVM_REGISTER_OP(_npi_minimum)
.set_attr("FCompute", BinaryBroadcastCompute);
+NNVM_REGISTER_OP(_npi_logaddexp2)
+.set_attr("FCompute", BinaryBroadcastCompute);
+
+NNVM_REGISTER_OP(_backward_logaddexp2)
+.set_attr("FCompute", BinaryBroadcastBackwardUseIn("FCompute",
BinaryBroadcastBackwardUseIn);
```
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on a change in pull request #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
reminisce commented on a change in pull request #15494: [Numpy] Added operator
logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15494#discussion_r302350456
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cu
##
@@ -78,5 +86,12 @@ NNVM_REGISTER_OP(_npi_maximum_scalar)
NNVM_REGISTER_OP(_npi_minimum_scalar)
.set_attr("FCompute", BinaryScalarOp::Compute);
+NNVM_REGISTER_OP(_npi_logaddexp2_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute);
+
+NNVM_REGISTER_OP(_backward_logaddexp2_scalar)
+.set_attr("FCompute", BinaryScalarOp::Backward
[GitHub] [incubator-mxnet] reminisce commented on a change in pull request #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
reminisce commented on a change in pull request #15494: [Numpy] Added operator
logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15494#discussion_r302350738
##
File path: src/operator/tensor/elemwise_binary_broadcast_op.h
##
@@ -580,17 +580,19 @@ inline void BinaryBroadcastBackwardUseInImpl(const
OpContext& ctx,
const TBlob ograd = inputs[0].reshape(new_oshape);
const TBlob lhs = inputs[1].reshape(new_lshape);
const TBlob rhs = inputs[2].reshape(new_rshape);
- size_t workspace_size_l = ReduceWorkspaceSize(
- s, lgrad.shape_, req[0], ograd.shape_, lhs.shape_, rhs.shape_);
- size_t workspace_size_r = ReduceWorkspaceSize(
- s, rgrad.shape_, req[1], ograd.shape_, lhs.shape_, rhs.shape_);
- size_t workspace_size = std::max(workspace_size_l, workspace_size_r);
- Tensor workspace =
- ctx.requested[0].get_space_typed(Shape1(workspace_size),
s);
- Reduce(s, lgrad, req[0],
workspace,
-ograd, lhs, rhs);
- Reduce(s, rgrad, req[1],
workspace,
-ograd, lhs, rhs);
+ if (inputs[0].shape_.Size() != 0) {
Review comment:
Instead of adding this condition check, it's better to return earlier if the
output gradient is a zero-size tensor.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] reminisce commented on issue #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
reminisce commented on issue #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn URL: https://github.com/apache/incubator-mxnet/pull/15494#issuecomment-510321190 Please rebase with the upstream numpy branch. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zixuanweeei commented on a change in pull request #15497: Independent gradients requests check with respect to weights and bias of convolution
zixuanweeei commented on a change in pull request #15497: Independent gradients
requests check with respect to weights and bias of convolution
URL: https://github.com/apache/incubator-mxnet/pull/15497#discussion_r302361443
##
File path: src/operator/nn/mkldnn/mkldnn_convolution.cc
##
@@ -662,21 +662,21 @@ void MKLDNNConvolutionBackward(const nnvm::NodeAttrs&
attrs, const OpContext &ct
in_grad[conv::kWeight],
convBwdWeight.bwdWeights_pd.diff_weights_primitive_desc(),
req[conv::kWeight]);
-mkldnn_output_t in_grad_bias;
-if (param.no_bias) {
- convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second);
- MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
-} else {
- in_grad_bias = CreateMKLDNNMem(
+
+if (!param.no_bias && req[conv::kBias]) {
+ auto in_grad_bias = CreateMKLDNNMem(
in_grad[conv::kBias],
convBwdWeight.bwdWeights_pd.diff_bias_primitive_desc(),
req[conv::kBias]);
convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second, *in_grad_bias.second);
+ *in_grad_weight.second, *in_grad_bias.second);
MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
CommitOutput(in_grad[conv::kBias], in_grad_bias);
+} else if (req[conv::kWeight]) {
+ convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
+ *in_grad_weight.second);
+ MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
}
-CommitOutput(in_grad[conv::kWeight], in_grad_weight);
+if (req[conv::kWeight]) CommitOutput(in_grad[conv::kWeight],
in_grad_weight);
Review comment:
It has the same behavior as req[kWeight]. Both of them return the operation
request type (`OpReqType`) to Forward and Backward. We can use it to control
the behavior of handling memory of result, like add/copy the result back to the
source memory or just do nothing with them.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] kshitij12345 commented on issue #15331: [fix] missing input log higher order.
kshitij12345 commented on issue #15331: [fix] missing input log higher order. URL: https://github.com/apache/incubator-mxnet/pull/15331#issuecomment-510326851 https://github.com/apache/incubator-mxnet/blob/5171e1d92cfc5eefa2c20dfe8ac3fac5351ad19a/src/operator/tensor/elemwise_unary_op_basic.cc#L1120 `dL/dygrad` for this one right? @larroy @apeforest , I was also wondering if we can check the number of inputs passed at compile time? I have observed the `MakeNode` gets the Op from dynamic registry based on the name. However we actually have information about the number of inputs and outputs for a given Op at compile time. I tried but couldn't actually figure out. What are your thoughts? How easy or hard would it be to check for valid number of inputs in `MakeNode`? This would help catch these sort of errors at compile time itself. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zixuanweeei commented on a change in pull request #15497: Independent gradients requests check with respect to weights and bias of convolution
zixuanweeei commented on a change in pull request #15497: Independent gradients
requests check with respect to weights and bias of convolution
URL: https://github.com/apache/incubator-mxnet/pull/15497#discussion_r302361443
##
File path: src/operator/nn/mkldnn/mkldnn_convolution.cc
##
@@ -662,21 +662,21 @@ void MKLDNNConvolutionBackward(const nnvm::NodeAttrs&
attrs, const OpContext &ct
in_grad[conv::kWeight],
convBwdWeight.bwdWeights_pd.diff_weights_primitive_desc(),
req[conv::kWeight]);
-mkldnn_output_t in_grad_bias;
-if (param.no_bias) {
- convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second);
- MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
-} else {
- in_grad_bias = CreateMKLDNNMem(
+
+if (!param.no_bias && req[conv::kBias]) {
+ auto in_grad_bias = CreateMKLDNNMem(
in_grad[conv::kBias],
convBwdWeight.bwdWeights_pd.diff_bias_primitive_desc(),
req[conv::kBias]);
convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
- *in_grad_weight.second, *in_grad_bias.second);
+ *in_grad_weight.second, *in_grad_bias.second);
MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
CommitOutput(in_grad[conv::kBias], in_grad_bias);
+} else if (req[conv::kWeight]) {
+ convBwdWeight.SetWeightNewMem(*data_mem, *out_grad_mem,
+ *in_grad_weight.second);
+ MKLDNNStream::Get()->RegisterPrim(convBwdWeight.GetBwdWeights());
}
-CommitOutput(in_grad[conv::kWeight], in_grad_weight);
+if (req[conv::kWeight]) CommitOutput(in_grad[conv::kWeight],
in_grad_weight);
Review comment:
It has the same behavior as req[kWeight]. Both of them return the operation
request type (OpReqType) to Forward and Backward. We can use it to control the
behavior of handling memory of result, like add/copy the result back to the
source memory or just do nothing with them.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] kshitij12345 commented on a change in pull request #15253: [MXNET-978] Add higher order gradient support `tan`, `tanh`
kshitij12345 commented on a change in pull request #15253: [MXNET-978] Add
higher order gradient support `tan`, `tanh`
URL: https://github.com/apache/incubator-mxnet/pull/15253#discussion_r302362521
##
File path: src/operator/tensor/elemwise_unary_op_trig.cc
##
@@ -139,7 +139,31 @@ The storage type of ``tan`` output depends upon the input
storage type:
)code" ADD_FILELINE)
.set_attr("FGradient", ElemwiseGradUseOut{ "_backward_tan" });
-MXNET_OPERATOR_REGISTER_BINARY_WITH_SPARSE_CPU_DR(_backward_tan,
unary_bwd);
+MXNET_OPERATOR_REGISTER_BINARY_WITH_SPARSE_CPU_DR(_backward_tan,
unary_bwd)
+.set_attr("FGradient",
+ [](const nnvm::NodePtr& n, const std::vector& ograds) {
+ // NodeEntry{n} : y_grad * f'(x)
+ // n->inputs[0] : y_grad
+ // n->inputs[1] : f(x) = tan(x)
+ // ograds[0] : head_grads
+ // f'(x) = sec^2(x)
+ // f''(x) = 2 * f'(x) * f(x)
+ const std::unordered_map args = {{"scalar",
"2.0"}};
+ auto two_y = MakeNode("_mul_scalar", n->attrs.name + "_mul_two",
{n->inputs[1]}, &args, &n);
+ auto grad_grad_mid = MakeNode("elemwise_mul", n->attrs.name +
"_grad_mul",
Review comment:
Makes sense. Thanks. Will get to it.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] gyshi commented on a change in pull request #15312: Numpy bitwise_xor operator
gyshi commented on a change in pull request #15312: Numpy bitwise_xor operator
URL: https://github.com/apache/incubator-mxnet/pull/15312#discussion_r302366555
##
File path: src/operator/numpy/np_elemwise_binary_op.cc
##
@@ -0,0 +1,55 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_elemwise_binary_op.cc
+ * \brief CPU Implementation of numpy-compatible bitwise_xor operators
+ */
+
+#include
+#include "../mshadow_op.h" // mshadow operations
+#include "../operator_common.h" // MakeZeroGradNodes
+#include "../tensor/elemwise_binary_op.h" // ElemwiseShape, ElemwiseType
+#include "../tensor/elemwise_binary_broadcast_op.h" // BinaryBroadcastCompute
+
+namespace mxnet {
+namespace op {
+
+NNVM_REGISTER_OP(_np_bitwise_xor)
+.set_num_inputs(2)
+.set_num_outputs(1)
+.set_attr("FInferShape", BinaryBroadcastShape)
+.set_attr("FInferType", ElemwiseType<2, 1>)
Review comment:
done
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] hgt312 opened a new pull request #15508: Numpy Tril (Lower triangle) operator
hgt312 opened a new pull request #15508: Numpy Tril (Lower triangle) operator URL: https://github.com/apache/incubator-mxnet/pull/15508 ## Description ## Implementation of numpy tril operator in mxnet. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - add `tril` op ## Comments ## Welcome @reminisce and others for reviewing. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ThomasDelteil commented on issue #15448: [MKLDNN]Enhance Quantization APIs and Tutorial
ThomasDelteil commented on issue #15448: [MKLDNN]Enhance Quantization APIs and Tutorial URL: https://github.com/apache/incubator-mxnet/pull/15448#issuecomment-510336674 Will do, at a conference this week, limited bandwidth but next week I'll have some availability to look into quantization again and get back to you on the different email threads as well, apologies for the delay! This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] yidawang commented on issue #15465: [RFC] Integrate TVM into Apache MXNet
yidawang commented on issue #15465: [RFC] Integrate TVM into Apache MXNet URL: https://github.com/apache/incubator-mxnet/issues/15465#issuecomment-510345628 > @yidawang If we all agree that we should avoid using more than one thread pool implementation in the runtime, then I guess OMP is the only choice, is that true? Sorry for the late reply, traveling these days. In this case, I think we should also benchmark to compare the performance of between different scenarios to decide. Theoretically, if a model runs both TVM ops and original MXNet ops on CPUs, I agree that using OpenMP may be a short term solution. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] Maicus closed issue #15484: Binding Model fails with simple_bind error
Maicus closed issue #15484: Binding Model fails with simple_bind error URL: https://github.com/apache/incubator-mxnet/issues/15484 This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] CynthiaProtector opened a new issue #15509: Train models under the director '/mxnet/example/gluon/'
CynthiaProtector opened a new issue #15509: Train models under the director '/mxnet/example/gluon/' URL: https://github.com/apache/incubator-mxnet/issues/15509 I conduct distributed training on almost all the models provided under the directory '/mxnet/example/gluon', however, when running the models with the asynchronous mechanism (--kvstore dist_async), none of the models is convergent. For instance, I train the resnet18_v1 with cifar10 dataset, the training accuracy is around 0.09 under the asynchronous mechanism. Nevertheless, training with the synchronous mechanism can achieve the target accuracy. I do not know the reason why the asynchronous SGD is not applicable to these models, and what should I do to fix this? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] zoeygxy commented on a change in pull request #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
zoeygxy commented on a change in pull request #15494: [Numpy] Added operator
logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15494#discussion_r302417072
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cu
##
@@ -78,5 +86,12 @@ NNVM_REGISTER_OP(_npi_maximum_scalar)
NNVM_REGISTER_OP(_npi_minimum_scalar)
.set_attr("FCompute", BinaryScalarOp::Compute);
+NNVM_REGISTER_OP(_npi_logaddexp2_scalar)
+.set_attr("FCompute", BinaryScalarOp::Compute);
+
+NNVM_REGISTER_OP(_backward_logaddexp2_scalar)
+.set_attr("FCompute", BinaryScalarOp::Backward
[GitHub] [incubator-mxnet] zoeygxy commented on a change in pull request #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
zoeygxy commented on a change in pull request #15494: [Numpy] Added operator
logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15494#discussion_r302417006
##
File path: src/operator/numpy/np_elemwise_broadcast_op.cu
##
@@ -48,6 +48,14 @@ NNVM_REGISTER_OP(_npi_maximum)
NNVM_REGISTER_OP(_npi_minimum)
.set_attr("FCompute", BinaryBroadcastCompute);
+NNVM_REGISTER_OP(_npi_logaddexp2)
+.set_attr("FCompute", BinaryBroadcastCompute);
+
+NNVM_REGISTER_OP(_backward_logaddexp2)
+.set_attr("FCompute", BinaryBroadcastBackwardUseIn
[GitHub] [incubator-mxnet] zoeygxy commented on a change in pull request #15494: [Numpy] Added operator logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
zoeygxy commented on a change in pull request #15494: [Numpy] Added operator
logaddexp2; added support for zero-size tensor in BinaryBroadcastBackwardUseIn
URL: https://github.com/apache/incubator-mxnet/pull/15494#discussion_r302417138
##
File path: src/operator/tensor/elemwise_binary_broadcast_op.h
##
@@ -580,17 +580,19 @@ inline void BinaryBroadcastBackwardUseInImpl(const
OpContext& ctx,
const TBlob ograd = inputs[0].reshape(new_oshape);
const TBlob lhs = inputs[1].reshape(new_lshape);
const TBlob rhs = inputs[2].reshape(new_rshape);
- size_t workspace_size_l = ReduceWorkspaceSize(
- s, lgrad.shape_, req[0], ograd.shape_, lhs.shape_, rhs.shape_);
- size_t workspace_size_r = ReduceWorkspaceSize(
- s, rgrad.shape_, req[1], ograd.shape_, lhs.shape_, rhs.shape_);
- size_t workspace_size = std::max(workspace_size_l, workspace_size_r);
- Tensor workspace =
- ctx.requested[0].get_space_typed(Shape1(workspace_size),
s);
- Reduce(s, lgrad, req[0],
workspace,
-ograd, lhs, rhs);
- Reduce(s, rgrad, req[1],
workspace,
-ograd, lhs, rhs);
+ if (inputs[0].shape_.Size() != 0) {
Review comment:
Will do, thanks!
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] gyshi opened a new pull request #15510: [Numpy] . Operator moveaxis
gyshi opened a new pull request #15510: [Numpy] . Operatormoveaxis URL: https://github.com/apache/incubator-mxnet/pull/15510 ## Description ## (Brief description on what this PR is about) ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [ ] Feature1, tests, (and when applicable, API doc) - [ ] Feature2, tests, (and when applicable, API doc) ## Comments ## - If this change is a backward incompatible change, why must this change be made. - Interesting edge cases to note here This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on a change in pull request #15449: cuda/cuDNN lib version checking. Force cuDNN v7 usage.
larroy commented on a change in pull request #15449: cuda/cuDNN lib version
checking. Force cuDNN v7 usage.
URL: https://github.com/apache/incubator-mxnet/pull/15449#discussion_r302428179
##
File path: src/common/cuda_utils.cc
##
@@ -0,0 +1,116 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * Copyright (c) 2019 by Contributors
+ * \file cuda_utils.cc
+ * \brief CUDA debugging utilities.
+ */
+
+#include
+#include "cuda_utils.h"
+
+#if MXNET_USE_CUDA == 1
+
+namespace mxnet {
+namespace common {
+namespace cuda {
+
+// The oldest version of cuda used in upstream MXNet CI testing, both for unix
and windows.
+// Users that have rebuilt MXNet against older versions will we advised with a
warning to upgrade
+// their systems to match the CI level. Minimally, users should rerun the CI
locally.
+#if defined(_MSC_VER)
+#define MXNET_CI_OLDEST_CUDA_VERSION 9020
+#else
+#define MXNET_CI_OLDEST_CUDA_VERSION 1
+#endif
+
+
+// Start-up check that the version of cuda compiled-against matches the
linked-against version.
+bool CudaVersionChecks() {
+ // Don't bother with checks if there are no GPUs visible (e.g. with
CUDA_VISIBLE_DEVICES="")
+ if (dmlc::GetEnv("MXNET_CUDA_VERSION_CHECKING", true) &&
Context::GetGPUCount() > 0) {
+int linkedAgainstCudaVersion = 0;
+CUDA_CALL(cudaRuntimeGetVersion(&linkedAgainstCudaVersion));
+if (linkedAgainstCudaVersion != CUDA_VERSION)
+ LOG(WARNING) << "cuda library mismatch: linked-against version " <<
linkedAgainstCudaVersion
+ << " != compiled-against version " << CUDA_VERSION << "."
+ << "Set MXNET_CUDA_VERSION_CHECKING=0 to quiet this
warning.";
+if (CUDA_VERSION < MXNET_CI_OLDEST_CUDA_VERSION)
+ LOG(WARNING) << "Upgrade advisory: this mxnet has been built against
cuda library version "
+ << CUDA_VERSION << ", which is older than the oldest
version tested by CI ("
+ << MXNET_CI_OLDEST_CUDA_VERSION << "). "
+ << "Set MXNET_CUDA_VERSION_CHECKING=0 to quiet this
warning.";
+ }
+ return true;
+}
+
+// Dynamic initialization here will emit a warning if runtime and compile-time
versions mismatch.
+// Also if the user has recompiled their source to a version no longer tested
by upstream CI.
+bool cuda_version_ok = CudaVersionChecks();
+
+} // namespace cuda
+} // namespace common
+} // namespace mxnet
+
+#endif // MXNET_USE_CUDA
+
+#if MXNET_USE_CUDNN == 1
+
+namespace mxnet {
+namespace common {
+namespace cudnn {
+
+// The oldest version of CUDNN used in upstream MXNet CI testing, both for
unix and windows.
+// Users that have rebuilt MXNet against older versions will we advised with a
warning to upgrade
+// their systems to match the CI level. Minimally, users should rerun the CI
locally.
+#if defined(_MSC_VER)
+#define MXNET_CI_OLDEST_CUDNN_VERSION 7600
+#else
+#define MXNET_CI_OLDEST_CUDNN_VERSION 7600
+#endif
+
+// Start-up check that the version of cudnn compiled-against matches the
linked-against version.
+// Also if the user has recompiled their source to a version no longer tested
by upstream CI.
+bool CuDNNVersionChecks() {
+ // Don't bother with checks if there are no GPUs visible (e.g. with
CUDA_VISIBLE_DEVICES="")
+ if (dmlc::GetEnv("MXNET_CUDNN_VERSION_CHECKING", true) &&
Context::GetGPUCount() > 0) {
+size_t linkedAgainstCudnnVersion = cudnnGetVersion();
+if (linkedAgainstCudnnVersion != CUDNN_VERSION)
+ LOG(WARNING) << "cuDNN library mismatch: linked-against version " <<
linkedAgainstCudnnVersion
+ << " != compiled-against version " << CUDNN_VERSION << ". "
+ << "Set MXNET_CUDNN_VERSION_CHECKING=0 to quiet this
warning.";
+if (CUDNN_VERSION < MXNET_CI_OLDEST_CUDNN_VERSION)
+ LOG(WARNING) << "Upgrade advisory: this mxnet has been built against
cuDNN library version "
+ << CUDNN_VERSION << ", which is older than the oldest
version tested by CI ("
+ << MXNET_CI_OLDEST_CUDNN_VERSION << "). "
+ << "Set MXNET_CUDNN_VERSION_CHECKING=0 to quiet this
warning.";
+ }
+ return true;
+}
+
+// Dynamic initializa
[GitHub] [incubator-mxnet] hgt312 opened a new pull request #15511: Numpy Identity operator
hgt312 opened a new pull request #15511: Numpy Identity operator URL: https://github.com/apache/incubator-mxnet/pull/15511 ## Description ## Implementation of numpy `identity` operator in mxnet. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - add op `identity` - fix some typos/mistakes in doc of `zeros` and `ones` ## Comments ## Welcome @reminisce and others for reviewing. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] wshuail opened a new issue #15512: possible bug in nd.gather_nd
wshuail opened a new issue #15512: possible bug in nd.gather_nd
URL: https://github.com/apache/incubator-mxnet/issues/15512
When I use nd.gather_nd with gpu, a unexpected error happened sometimes.
Generally it may raise error like RuntimeError: cuda runtime error (77) : an
illegal memory access was encountered.
An error "terminate called without an active exception" could happen with
the code below.
`import os
import sys
import mxnet as mx
from mxnet import nd
sys.path.insert(0, os.path.expanduser('~/gluon_detector'))
ctx = mx.gpu(0)
for i in range(1):
if i % 100 == 0:
print (i)
batch_size = 2
num_classes = 2
width, height = 5, 5
k = 3
hm_origin =
nd.random.uniform(0, 10, (batch_size, num_classes, width, height), ctx=ctx)
hm = nd.reshape(hm_origin, (0, 0, -1))
topk_scores, topk_idx = nd.topk(hm, k=k, ret_typ='both')
topk_x_idx = nd.floor(topk_idx/width)
topk_y_idx = (topk_idx%height)
batch_idx = nd.repeat(nd.arange(batch_size),
repeats=num_classes*k).reshape((1, -1))
batch_idx = batch_idx.as_in_context(ctx)
class_idx = nd.repeat(nd.arange(num_classes),
repeats=batch_size*k).reshape((1, -1))
class_idx = class_idx.as_in_context(ctx)
topk_x_idx = nd.reshape(topk_x_idx, (1, -1))
topk_y_idx = nd.reshape(topk_y_idx, (1, -1))
indices = nd.concat(batch_idx, class_idx, topk_x_idx, topk_y_idx, dim=0)
results = nd.gather_nd(hm_origin, indices)
results = nd.reshape(results, (batch_size, num_classes, k))
`
when I add nd.waitall() at last, it works well.
any suggestions?
BTW, does mxnet has a function like torch.gather? gather_nd works close,
but not so convenient.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15512: possible bug in nd.gather_nd
mxnet-label-bot commented on issue #15512: possible bug in nd.gather_nd URL: https://github.com/apache/incubator-mxnet/issues/15512#issuecomment-510399079 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. Here are my recommended labels: Bug This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] wkcn opened a new pull request #15513: [WIP] add contrib op in cpp-pacakge
wkcn opened a new pull request #15513: [WIP] add contrib op in cpp-pacakge URL: https://github.com/apache/incubator-mxnet/pull/15513 ## Description ## Hi, there. I add the contrib operators into cpp-package. Currently, `dim_t (int64_t)` could not be parsed. I'm waiting for the merge of [this PR dmlc-core#540](https://github.com/dmlc/dmlc-core/pull/540). ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [ ] Changes are complete (i.e. I finished coding on this PR) - [ ] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [ ] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [ ] Feature1, tests, (and when applicable, API doc) - [ ] Feature2, tests, (and when applicable, API doc) ## Comments ## - If this change is a backward incompatible change, why must this change be made. - Interesting edge cases to note here This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] lebeg commented on issue #15452: fix nightly CI failure
lebeg commented on issue #15452: fix nightly CI failure URL: https://github.com/apache/incubator-mxnet/pull/15452#issuecomment-510455260 @roywei could you apply changes that @szha requested? The nightly tests are still broken. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] qqaatw commented on issue #15484: Binding Model fails with simple_bind error
qqaatw commented on issue #15484: Binding Model fails with simple_bind error URL: https://github.com/apache/incubator-mxnet/issues/15484#issuecomment-510480630 I have the same problem as you, could you tell more in detail about the Target and the TargetCode Section? thanks. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] gyshi commented on a change in pull request #15312: Numpy bitwise_xor operator
gyshi commented on a change in pull request #15312: Numpy bitwise_xor operator
URL: https://github.com/apache/incubator-mxnet/pull/15312#discussion_r302538899
##
File path: src/operator/numpy/np_elemwise_binary_op.cc
##
@@ -0,0 +1,55 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * \file np_elemwise_binary_op.cc
+ * \brief CPU Implementation of numpy-compatible bitwise_xor operators
+ */
+
+#include
+#include "../mshadow_op.h" // mshadow operations
+#include "../operator_common.h" // MakeZeroGradNodes
+#include "../tensor/elemwise_binary_op.h" // ElemwiseShape, ElemwiseType
+#include "../tensor/elemwise_binary_broadcast_op.h" // BinaryBroadcastCompute
+
+namespace mxnet {
+namespace op {
+
+NNVM_REGISTER_OP(_np_bitwise_xor)
+.set_num_inputs(2)
+.set_num_outputs(1)
+.set_attr("FInferShape", BinaryBroadcastShape)
+.set_attr("FInferType", ElemwiseType<2, 1>)
Review comment:
resolved, it's my first op, i donot use npx.set_np(), and i attempt again,
in mxnet, bool will be changed to 1.0 or 0.0, so it can support bool.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] marcoabreu commented on issue #15506: Improving error message
marcoabreu commented on issue #15506: Improving error message URL: https://github.com/apache/incubator-mxnet/issues/15506#issuecomment-510483367 ```No space left on device``` Sounds pretty clear to me. Is this not talking about disk space? What was the root cause? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] smissan opened a new issue #15514: dmlc.lib is not generated
smissan opened a new issue #15514: dmlc.lib is not generated URL: https://github.com/apache/incubator-mxnet/issues/15514 when trying to build latest head, dmlc.dll is being generated, but not dmlc.lib (using ms visual studio 2015). Therefore, building mxnet dll fails with 1>-- Build started: Project: mxnet, Configuration: Release x64 -- 1>LINK : fatal error LNK1181: cannot open input file '3rdparty\dmlc-core\Release\dmlc.lib' This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] mxnet-label-bot commented on issue #15514: dmlc.lib is not generated
mxnet-label-bot commented on issue #15514: dmlc.lib is not generated URL: https://github.com/apache/incubator-mxnet/issues/15514#issuecomment-510484008 Hey, this is the MXNet Label Bot. Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it. Here are my recommended labels: Build This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] kshitij12345 opened a new pull request #15515: [MXNET-978] Higher Order Gradient Support `arcsin`, `arccos`.
kshitij12345 opened a new pull request #15515: [MXNET-978] Higher Order Gradient Support `arcsin`, `arccos`. URL: https://github.com/apache/incubator-mxnet/pull/15515 ## Description ## PR intends to add support for higher order gradient for `arcsin`, `arccos`. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [x] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA-978 issue](https://issues.apache.org/jira/browse/MXNET-978) created (except PRs with tiny changes) - [x] Changes are complete (i.e. I finished coding on this PR) - [x] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [x] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [x] higher order gradient for a `arcsin`, `arccos`. - [x] unit test for the same. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] qqaatw edited a comment on issue #15484: Binding Model fails with simple_bind error
qqaatw edited a comment on issue #15484: Binding Model fails with simple_bind error URL: https://github.com/apache/incubator-mxnet/issues/15484#issuecomment-510480630 @Maicus I have the same problem as you, could you tell more in detail about the Target and the TargetCode Section? thanks. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] aaronmarkham commented on a change in pull request #15454: Julia docs
aaronmarkham commented on a change in pull request #15454: Julia docs
URL: https://github.com/apache/incubator-mxnet/pull/15454#discussion_r302618951
##
File path: docs/install/ubuntu_setup.md
##
@@ -310,25 +310,93 @@ Refer to the [Clojure setup
guide](https://github.com/apache/incubator-mxnet/tre
### Install the MXNet Package for Julia
-The MXNet package for Julia is hosted in a separate repository, MXNet.jl,
which is available on [GitHub](https://github.com/dmlc/MXNet.jl). To use Julia
binding it with an existing libmxnet installation, set the ```MXNET_HOME```
environment variable by running the following command:
+ Install Julia
+The package available through `apt-get` is old and not compatible with the
latest version of MXNet.
+Fetch the latest version (1.0.3 at the time of this writing).
```bash
-export MXNET_HOME=//libmxnet
+wget -qO julia-10.tar.gz
https://julialang-s3.julialang.org/bin/linux/x64/1.0/julia-1.0.3-linux-x86_64.tar.gz
```
-The path to the existing libmxnet installation should be the root directory of
libmxnet. In other words, you should be able to find the ```libmxnet.so``` file
at ```$MXNET_HOME/lib```. For example, if the root directory of libmxnet is
```~```, you would run the following command:
+Place the extracted files somewhere like a julia folder in your home dir.
```bash
-export MXNET_HOME=/~/libmxnet
+mkdir ~/julia
+mv julia-10.tar.gz ~/julia
+cd ~/julia
+tar xvf julia-10.tar.gz
```
-You might want to add this command to your ```~/.bashrc``` file. If you do,
you can install the Julia package in the Julia console using the following
command:
+Test Julia.
+```bash
+cd julia-1.0.3/bin
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+If you're still getting the old version, remove it.
+```bash
+sudo apt remove julia
+```
+
+Update your PATH to have Julia's new location. Add this to your `.zshrc`,
`.bashrc`, `.profile` or `.bash_profile`.
+```bash
+export PATH=~/julia/julia-1.0.3/bin:$PATH
+```
+
+Validate your PATH.
+```bash
+echo $PATH
+```
+
+Validate Julia works and is the expected version.
+```bash
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+ Setup Your MXNet-Julia Environment
+
+**For each of the following environment variables, add the commands to your
`.zshrc`, `.bashrc`, `.profile` or `.bash_profile` to make them persist.**
+
+Create a `julia-depot` folder and environment variable.
+```bash
+mkdir julia-depot
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+```
+
+To use the Julia binding with an existing `libmxnet` installation, set the
`MXNET_HOME` environment variable to the MXNet source root. For example:
+```bash
+export MXNET_HOME=$HOME/incubator-mxnet
+```
-```julia
-Pkg.add("MXNet")
+Now set the `LD_LIBRARY_PATH` environment variable to where `libmxnet.so` is
found. If you can't find it, you might have skipped the building MXNet step. Go
back and [build MXNet](#build-the-shared-library) first. For example:
+```bash
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+```
+
+Verify the location of `libjemalloc.so` and set the `LD_PRELOAD` environment
variable.
+```bash
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+With all of these updates, here's an example of what you might want to have in
your `.zshrc`, `.bashrc`, `.profile` or `.bash_profile`.
+
+```
+export PATH=$HOME/bin:$HOME/.local/bin:$HOME/julia/julia-1.0.3/bin:$PATH
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+export MXNET_HOME=$HOME/incubator-mxnet
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+Install MXNet with Julia:
+
+```bash
+julia --color=yes --project=./ -e \
+ 'using Pkg; \
+ Pkg.develop(PackageSpec(name="MXNet", path =
joinpath(ENV["MXNET_HOME"], "julia")))'
Review comment:
Yes, let's update things in another PR when 1.5 comes out.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] aaronmarkham commented on a change in pull request #15454: Julia docs
aaronmarkham commented on a change in pull request #15454: Julia docs
URL: https://github.com/apache/incubator-mxnet/pull/15454#discussion_r302619949
##
File path: docs/install/ubuntu_setup.md
##
@@ -310,25 +310,93 @@ Refer to the [Clojure setup
guide](https://github.com/apache/incubator-mxnet/tre
### Install the MXNet Package for Julia
-The MXNet package for Julia is hosted in a separate repository, MXNet.jl,
which is available on [GitHub](https://github.com/dmlc/MXNet.jl). To use Julia
binding it with an existing libmxnet installation, set the ```MXNET_HOME```
environment variable by running the following command:
+ Install Julia
+The package available through `apt-get` is old and not compatible with the
latest version of MXNet.
+Fetch the latest version (1.0.3 at the time of this writing).
```bash
-export MXNET_HOME=//libmxnet
+wget -qO julia-10.tar.gz
https://julialang-s3.julialang.org/bin/linux/x64/1.0/julia-1.0.3-linux-x86_64.tar.gz
```
-The path to the existing libmxnet installation should be the root directory of
libmxnet. In other words, you should be able to find the ```libmxnet.so``` file
at ```$MXNET_HOME/lib```. For example, if the root directory of libmxnet is
```~```, you would run the following command:
+Place the extracted files somewhere like a julia folder in your home dir.
```bash
-export MXNET_HOME=/~/libmxnet
+mkdir ~/julia
+mv julia-10.tar.gz ~/julia
+cd ~/julia
+tar xvf julia-10.tar.gz
```
-You might want to add this command to your ```~/.bashrc``` file. If you do,
you can install the Julia package in the Julia console using the following
command:
+Test Julia.
+```bash
+cd julia-1.0.3/bin
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+If you're still getting the old version, remove it.
+```bash
+sudo apt remove julia
+```
+
+Update your PATH to have Julia's new location. Add this to your `.zshrc`,
`.bashrc`, `.profile` or `.bash_profile`.
+```bash
+export PATH=~/julia/julia-1.0.3/bin:$PATH
+```
+
+Validate your PATH.
+```bash
+echo $PATH
+```
+
+Validate Julia works and is the expected version.
+```bash
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+ Setup Your MXNet-Julia Environment
+
+**For each of the following environment variables, add the commands to your
`.zshrc`, `.bashrc`, `.profile` or `.bash_profile` to make them persist.**
+
+Create a `julia-depot` folder and environment variable.
+```bash
+mkdir julia-depot
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+```
+
+To use the Julia binding with an existing `libmxnet` installation, set the
`MXNET_HOME` environment variable to the MXNet source root. For example:
+```bash
+export MXNET_HOME=$HOME/incubator-mxnet
+```
-```julia
-Pkg.add("MXNet")
+Now set the `LD_LIBRARY_PATH` environment variable to where `libmxnet.so` is
found. If you can't find it, you might have skipped the building MXNet step. Go
back and [build MXNet](#build-the-shared-library) first. For example:
+```bash
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+```
+
+Verify the location of `libjemalloc.so` and set the `LD_PRELOAD` environment
variable.
+```bash
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+With all of these updates, here's an example of what you might want to have in
your `.zshrc`, `.bashrc`, `.profile` or `.bash_profile`.
+
+```
+export PATH=$HOME/bin:$HOME/.local/bin:$HOME/julia/julia-1.0.3/bin:$PATH
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
Review comment:
Ok, since it doesn't break anything but is unnecessary let's remove it in
the next PR. That way I don't have to shepherd CI again.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] iblis17 commented on a change in pull request #15454: Julia docs
iblis17 commented on a change in pull request #15454: Julia docs
URL: https://github.com/apache/incubator-mxnet/pull/15454#discussion_r302620627
##
File path: docs/install/ubuntu_setup.md
##
@@ -310,25 +310,93 @@ Refer to the [Clojure setup
guide](https://github.com/apache/incubator-mxnet/tre
### Install the MXNet Package for Julia
-The MXNet package for Julia is hosted in a separate repository, MXNet.jl,
which is available on [GitHub](https://github.com/dmlc/MXNet.jl). To use Julia
binding it with an existing libmxnet installation, set the ```MXNET_HOME```
environment variable by running the following command:
+ Install Julia
+The package available through `apt-get` is old and not compatible with the
latest version of MXNet.
+Fetch the latest version (1.0.3 at the time of this writing).
```bash
-export MXNET_HOME=//libmxnet
+wget -qO julia-10.tar.gz
https://julialang-s3.julialang.org/bin/linux/x64/1.0/julia-1.0.3-linux-x86_64.tar.gz
```
-The path to the existing libmxnet installation should be the root directory of
libmxnet. In other words, you should be able to find the ```libmxnet.so``` file
at ```$MXNET_HOME/lib```. For example, if the root directory of libmxnet is
```~```, you would run the following command:
+Place the extracted files somewhere like a julia folder in your home dir.
```bash
-export MXNET_HOME=/~/libmxnet
+mkdir ~/julia
+mv julia-10.tar.gz ~/julia
+cd ~/julia
+tar xvf julia-10.tar.gz
```
-You might want to add this command to your ```~/.bashrc``` file. If you do,
you can install the Julia package in the Julia console using the following
command:
+Test Julia.
+```bash
+cd julia-1.0.3/bin
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+If you're still getting the old version, remove it.
+```bash
+sudo apt remove julia
+```
+
+Update your PATH to have Julia's new location. Add this to your `.zshrc`,
`.bashrc`, `.profile` or `.bash_profile`.
+```bash
+export PATH=~/julia/julia-1.0.3/bin:$PATH
+```
+
+Validate your PATH.
+```bash
+echo $PATH
+```
+
+Validate Julia works and is the expected version.
+```bash
+julia -e 'using InteractiveUtils; versioninfo()'
+```
+
+ Setup Your MXNet-Julia Environment
+
+**For each of the following environment variables, add the commands to your
`.zshrc`, `.bashrc`, `.profile` or `.bash_profile` to make them persist.**
+
+Create a `julia-depot` folder and environment variable.
+```bash
+mkdir julia-depot
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+```
+
+To use the Julia binding with an existing `libmxnet` installation, set the
`MXNET_HOME` environment variable to the MXNet source root. For example:
+```bash
+export MXNET_HOME=$HOME/incubator-mxnet
+```
-```julia
-Pkg.add("MXNet")
+Now set the `LD_LIBRARY_PATH` environment variable to where `libmxnet.so` is
found. If you can't find it, you might have skipped the building MXNet step. Go
back and [build MXNet](#build-the-shared-library) first. For example:
+```bash
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+```
+
+Verify the location of `libjemalloc.so` and set the `LD_PRELOAD` environment
variable.
+```bash
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+With all of these updates, here's an example of what you might want to have in
your `.zshrc`, `.bashrc`, `.profile` or `.bash_profile`.
+
+```
+export PATH=$HOME/bin:$HOME/.local/bin:$HOME/julia/julia-1.0.3/bin:$PATH
+export JULIA_DEPOT_PATH=$HOME/julia/julia-depot
+export MXNET_HOME=$HOME/incubator-mxnet
+export LD_LIBRARY_PATH=$HOME/incubator-mxnet/lib:$LD_LIBRARY_PATH
+export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libjemalloc.so:$LD_PRELOAD
+```
+
+Install MXNet with Julia:
+
+```bash
+julia --color=yes --project=./ -e \
+ 'using Pkg; \
+ Pkg.develop(PackageSpec(name="MXNet", path =
joinpath(ENV["MXNET_HOME"], "julia")))'
Review comment:
ah, :ok_hand:
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] iblis17 commented on issue #15454: Julia docs
iblis17 commented on issue #15454: Julia docs URL: https://github.com/apache/incubator-mxnet/pull/15454#issuecomment-510545691 Nice work! @aaronmarkham This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] access2rohit commented on issue #15490: Utility to help developers debug operators: Tensor Inspector
access2rohit commented on issue #15490: Utility to help developers debug operators: Tensor Inspector URL: https://github.com/apache/incubator-mxnet/pull/15490#issuecomment-510561397 Can you split the PR into 2 separate PRs. One for CPU and one for GPU ? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] matteosal commented on issue #15497: [MKLDNN] Independent gradients requests check with respect to weights and bias of convolution
matteosal commented on issue #15497: [MKLDNN] Independent gradients requests
check with respect to weights and bias of convolution
URL: https://github.com/apache/incubator-mxnet/pull/15497#issuecomment-510563332
The example of https://github.com/apache/incubator-mxnet/issues/15464 is
fixed here, but I see a failure with this one, where the weights gradient is
requested in isolation (so the opposite of
https://github.com/apache/incubator-mxnet/issues/15464 ):
```
import mxnet as mx
sym = mx.sym.Convolution(
mx.sym.Variable('in'),
mx.sym.Variable('w'),
mx.sym.Variable('b'),
kernel=(1, 1),
num_filter=1
)
args = {
'in': mx.nd.ones([1, 1, 3, 3]),
'w': mx.nd.ones([1, 1, 1, 1]),
'b': mx.nd.ones([1]),
}
grad = {
'in': mx.nd.zeros([1, 1, 3, 3]),
'w': mx.nd.zeros([1, 1, 1, 1]),
'b': mx.nd.zeros([1]),
}
req = {'in': 'null', 'w': 'write', 'b': 'null'}
outgrad = mx.nd.ones([1, 1, 3, 3])
ex = sym.bind(mx.cpu(), args, args_grad=grad, grad_req=req)
ex.forward(True);
ex.backward(out_grads=outgrad);
mx.ndarray.waitall()
```
This is what gets printed to command line:
```
Traceback (most recent call last):
File "script2.py", line 27, in
mx.ndarray.waitall()
File "/home/matteo/Git/mxnet/python/mxnet/ndarray/ndarray.py", line 166,
in waitall
check_call(_LIB.MXNDArrayWaitAll())
File "/home/matteo/Git/mxnet/python/mxnet/base.py", line 253, in check_call
raise MXNetError(py_str(_LIB.MXGetLastError()))
mxnet.base.MXNetError: std::exception
```
It doesn't fail on master
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] kshitij12345 commented on a change in pull request #15253: [MXNET-978] Add higher order gradient support `tan`, `tanh`
kshitij12345 commented on a change in pull request #15253: [MXNET-978] Add
higher order gradient support `tan`, `tanh`
URL: https://github.com/apache/incubator-mxnet/pull/15253#discussion_r30264
##
File path: src/operator/tensor/elemwise_unary_op_trig.cc
##
@@ -139,7 +139,31 @@ The storage type of ``tan`` output depends upon the input
storage type:
)code" ADD_FILELINE)
.set_attr("FGradient", ElemwiseGradUseOut{ "_backward_tan" });
-MXNET_OPERATOR_REGISTER_BINARY_WITH_SPARSE_CPU_DR(_backward_tan,
unary_bwd);
+MXNET_OPERATOR_REGISTER_BINARY_WITH_SPARSE_CPU_DR(_backward_tan,
unary_bwd)
+.set_attr("FGradient",
+ [](const nnvm::NodePtr& n, const std::vector& ograds) {
+ // NodeEntry{n} : y_grad * f'(x)
+ // n->inputs[0] : y_grad
+ // n->inputs[1] : f(x) = tan(x)
+ // ograds[0] : head_grads
+ // f'(x) = sec^2(x)
+ // f''(x) = 2 * f'(x) * f(x)
+ const std::unordered_map args = {{"scalar",
"2.0"}};
+ auto two_y = MakeNode("_mul_scalar", n->attrs.name + "_mul_two",
{n->inputs[1]}, &args, &n);
+ auto grad_grad_mid = MakeNode("elemwise_mul", n->attrs.name +
"_grad_mul",
Review comment:
I have updated the comment. See if it is okay? Or maybe the phrasing can be
improved.
Thanks
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] frankfliu commented on issue #15509: Train models under the director '/mxnet/example/gluon/'
frankfliu commented on issue #15509: Train models under the director '/mxnet/example/gluon/' URL: https://github.com/apache/incubator-mxnet/issues/15509#issuecomment-510566851 @mxnet-label-bot add [Question, distributed, training] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] frankfliu commented on issue #15514: dmlc.lib is not generated
frankfliu commented on issue #15514: dmlc.lib is not generated URL: https://github.com/apache/incubator-mxnet/issues/15514#issuecomment-510568547 @mxnet-label-bot add [build] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] frankfliu commented on issue #15512: possible bug in nd.gather_nd
frankfliu commented on issue #15512: possible bug in nd.gather_nd URL: https://github.com/apache/incubator-mxnet/issues/15512#issuecomment-510568295 @mxnet-label-bot add [backend, cuda, question] This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #15506: Improving error message
ChaiBapchya commented on issue #15506: Improving error message URL: https://github.com/apache/incubator-mxnet/issues/15506#issuecomment-510574285 I had an issue with this part of that error ` when writing output to : ` And I am running this on p2.8x large GPU-optimized AWS EC2 instance with EBS of 100gig But somehow falls into the space issue I'm surely missing something. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] DickJC123 commented on a change in pull request #15449: cuda/cuDNN lib version checking. Force cuDNN v7 usage.
DickJC123 commented on a change in pull request #15449: cuda/cuDNN lib version
checking. Force cuDNN v7 usage.
URL: https://github.com/apache/incubator-mxnet/pull/15449#discussion_r302657800
##
File path: src/common/cuda_utils.cc
##
@@ -0,0 +1,108 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * Copyright (c) 2019 by Contributors
+ * \file cuda_utils.cc
+ * \brief CUDA debugging utilities.
+ */
+
+#include
+#include "cuda_utils.h"
+
+#if MXNET_USE_CUDA == 1
+
+namespace mxnet {
+namespace common {
+namespace cuda {
+
+// The oldest version of cuda used in upstream MXNet CI testing, both for unix
and windows.
+// Users that have rebuilt MXNet against older versions will we advised with a
warning to upgrade
+// their systems to match the CI level. Minimally, users should rerun the CI
locally.
+#if defined(_MSC_VER)
+#define MXNET_CI_OLDEST_CUDA_VERSION 9020
+#else
+#define MXNET_CI_OLDEST_CUDA_VERSION 1
+#endif
+
+// Dynamic init here will emit a warning if runtime and compile-time cuda lib
versions mismatch.
+// Also if the user has recompiled their source to a version no longer tested
by upstream CI.
+bool cuda_version_check_performed = []() {
+ // Don't bother with checks if there are no GPUs visible (e.g. with
CUDA_VISIBLE_DEVICES="")
+ if (dmlc::GetEnv("MXNET_CUDA_VERSION_CHECKING", true) &&
Context::GetGPUCount() > 0) {
+int linkedAgainstCudaVersion = 0;
+CUDA_CALL(cudaRuntimeGetVersion(&linkedAgainstCudaVersion));
+if (linkedAgainstCudaVersion != CUDA_VERSION)
+ LOG(WARNING) << "cuda library mismatch: linked-against version " <<
linkedAgainstCudaVersion
Review comment:
So 'yes' there would be a warning if the user built against 10.1, but ran
with 10.2. These warnings can be turned off with an environment variable
setting MXNET_CUDA_VERSION_CHECKING=0. The idea behind the 'advisory' is that
the user may want to rebuild to get the new functionality present in 10.2, or
perhaps to avoid work-arounds for any issues of 10.1. It's probably more
useful with the CUDNN version checks, where we have far more compile guards
based on version minor numbers. Do you feel these warnings would be unwelcome
to users?
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya commented on issue #15506: Improving error message
ChaiBapchya commented on issue #15506: Improving error message URL: https://github.com/apache/incubator-mxnet/issues/15506#issuecomment-510581504 Issue resolved by 1. Switching to Ubuntu AMI (instead of DL AMI that comes with loads of other packages and libraries I don't need) 2. Increasing the root storage capacity (default 8gig to 20) @marcoabreu coming to this specific issue - "when writing output to " seems abrupt. We can either do away with that or just mention the location where was the process writing output to. What do you reckon? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] ChaiBapchya edited a comment on issue #15506: Improving error message
ChaiBapchya edited a comment on issue #15506: Improving error message URL: https://github.com/apache/incubator-mxnet/issues/15506#issuecomment-510581504 So you're right. Root cause was related to disk space. (Though it is weird how even after mounting 100gig of EBS volume on /data directory, it ran into this?) Issue resolved by 1. Switching to Ubuntu AMI (instead of DL AMI that comes with loads of other packages and libraries I don't need) 2. Increasing the root storage capacity (default 8gig to 20) @marcoabreu coming to this specific issue - "when writing output to " seems abrupt. We can either do away with that or just mention the location where was the process writing output to. What do you reckon? This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] kevinzh92 opened a new pull request #15516: Fix memory leak reported by ASAN in NNVM to ONNX conversion
kevinzh92 opened a new pull request #15516: Fix memory leak reported by ASAN in NNVM to ONNX conversion URL: https://github.com/apache/incubator-mxnet/pull/15516 ## Description ## Fix memory leak reported by ASAN in NNVM to ONNX conversion of a constant. ## Checklist ## ### Essentials ### Please feel free to remove inapplicable items for your PR. - [ ] The PR title starts with [MXNET-$JIRA_ID], where $JIRA_ID refers to the relevant [JIRA issue](https://issues.apache.org/jira/projects/MXNET/issues) created (except PRs with tiny changes) - [x] Changes are complete (i.e. I finished coding on this PR) - [x] All changes have test coverage: - Unit tests are added for small changes to verify correctness (e.g. adding a new operator) - Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore) - Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL) - [ ] Code is well-documented: - For user-facing API changes, API doc string has been updated. - For new C++ functions in header files, their functionalities and arguments are documented. - For new examples, README.md is added to explain the what the example does, the source of the dataset, expected performance on test set and reference to the original paper if applicable - Check the API doc at http://mxnet-ci-doc.s3-accelerate.dualstack.amazonaws.com/PR-$PR_ID/$BUILD_ID/index.html - [x] To the my best knowledge, examples are either not affected by this change, or have been fixed to be compatible with this change ### Changes ### - [x] Fixed a memory leak in NNVM to ONNX constants conversion code. The `shared_ptr`, as written, does not call `delete[]` on a dynamically allocated array. ## Comments ## This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on a change in pull request #15449: cuda/cuDNN lib version checking. Force cuDNN v7 usage.
larroy commented on a change in pull request #15449: cuda/cuDNN lib version
checking. Force cuDNN v7 usage.
URL: https://github.com/apache/incubator-mxnet/pull/15449#discussion_r302695533
##
File path: src/common/cuda_utils.cc
##
@@ -0,0 +1,108 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied. See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+
+/*!
+ * Copyright (c) 2019 by Contributors
+ * \file cuda_utils.cc
+ * \brief CUDA debugging utilities.
+ */
+
+#include
+#include "cuda_utils.h"
+
+#if MXNET_USE_CUDA == 1
+
+namespace mxnet {
+namespace common {
+namespace cuda {
+
+// The oldest version of cuda used in upstream MXNet CI testing, both for unix
and windows.
+// Users that have rebuilt MXNet against older versions will we advised with a
warning to upgrade
+// their systems to match the CI level. Minimally, users should rerun the CI
locally.
+#if defined(_MSC_VER)
+#define MXNET_CI_OLDEST_CUDA_VERSION 9020
+#else
+#define MXNET_CI_OLDEST_CUDA_VERSION 1
+#endif
+
+// Dynamic init here will emit a warning if runtime and compile-time cuda lib
versions mismatch.
+// Also if the user has recompiled their source to a version no longer tested
by upstream CI.
+bool cuda_version_check_performed = []() {
+ // Don't bother with checks if there are no GPUs visible (e.g. with
CUDA_VISIBLE_DEVICES="")
+ if (dmlc::GetEnv("MXNET_CUDA_VERSION_CHECKING", true) &&
Context::GetGPUCount() > 0) {
+int linkedAgainstCudaVersion = 0;
+CUDA_CALL(cudaRuntimeGetVersion(&linkedAgainstCudaVersion));
+if (linkedAgainstCudaVersion != CUDA_VERSION)
+ LOG(WARNING) << "cuda library mismatch: linked-against version " <<
linkedAgainstCudaVersion
Review comment:
I think the question is what can be the issues when linking against a
smaller cuda, leaving performance gains on the table? I think you guys are the
experts, I was getting some info from here:
https://docs.nvidia.com/deploy/cuda-compatibility/#binary-compatibility
Does this warning indicate a real problem or will it confuse users, when
there's nothing wrong on running with a newer cuda.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
[GitHub] [incubator-mxnet] larroy edited a comment on issue #15285: Graph dumper
larroy edited a comment on issue #15285: Graph dumper URL: https://github.com/apache/incubator-mxnet/pull/15285#issuecomment-509858209 @ptrendx Thanks I will try to use your suggestion in the next iteration. This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
[GitHub] [incubator-mxnet] larroy commented on a change in pull request #15253: [MXNET-978] Add higher order gradient support `tan`, `tanh`
larroy commented on a change in pull request #15253: [MXNET-978] Add higher
order gradient support `tan`, `tanh`
URL: https://github.com/apache/incubator-mxnet/pull/15253#discussion_r302708484
##
File path: src/operator/tensor/elemwise_unary_op_trig.cc
##
@@ -139,7 +139,31 @@ The storage type of ``tan`` output depends upon the input
storage type:
)code" ADD_FILELINE)
.set_attr("FGradient", ElemwiseGradUseOut{ "_backward_tan" });
-MXNET_OPERATOR_REGISTER_BINARY_WITH_SPARSE_CPU_DR(_backward_tan,
unary_bwd);
+MXNET_OPERATOR_REGISTER_BINARY_WITH_SPARSE_CPU_DR(_backward_tan,
unary_bwd)
+.set_attr("FGradient",
+ [](const nnvm::NodePtr& n, const std::vector& ograds) {
+ // NodeEntry{n} : y_grad * f'(x)
+ // n->inputs[0] : y_grad
+ // n->inputs[1] : f(x) = tan(x)
+ // ograds[0] : head_grads
+ // f'(x) = sec^2(x)
+ // f''(x) = 2 * f'(x) * f(x)
+ const std::unordered_map args = {{"scalar",
"2.0"}};
+ auto two_y = MakeNode("_mul_scalar", n->attrs.name + "_mul_two",
{n->inputs[1]}, &args, &n);
+ auto grad_grad_mid = MakeNode("elemwise_mul", n->attrs.name +
"_grad_mul",
Review comment:
Thanks. About the outputs, I think we should write some documentation
explaining what we are doing as I find it non trivial. Can you help me
understand the y_grad_grad (first output)?
If you want, we can move the conversation to the dev list or slack, as the
PR LGTM.
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
